| { | |
| "_name_or_path": "beam-searchers/test", | |
| "architectures": [ | |
| "LlamaForCausalLM" | |
| ], | |
| "attention_bias": false, | |
| "attention_dropout": 0.0, | |
| "bos_token_id": 128000, | |
| "eos_token_id": 128009, | |
| "hidden_act": "silu", | |
| "hidden_size": 4096, | |
| "initializer_range": 0.02, | |
| "intermediate_size": 14336, | |
| "max_position_embeddings": 8192, | |
| "model_type": "llama", | |
| "num_attention_heads": 32, | |
| "num_hidden_layers": 32, | |
| "num_key_value_heads": 8, | |
| "pretraining_tp": 1, | |
| "quantization_config": { | |
| "batch_size": 1, | |
| "bits": 4, | |
| "block_name_to_quantize": null, | |
| "cache_block_outputs": true, | |
| "damp_percent": 0.1, | |
| "dataset": [ | |
| "Question: Using a 4-gram character model, comparing \"banana\" and \"ananas\"...\n\nOptions:\nA. is the same as comparing P(ban)/P(an) to P(nas)/P(na)\nB. is the same as comparing P(bana) to P(anan)\nC. is the same as comparing \"aaabnn\" to \"aaanns\"\nD. is the same as comparing P(bana) to P(anas)\n\nAnswer:", | |
| "Question: Consider the exhaustive search of a uniformly distributed key in a set of size $N$. Think of the possible strategies and their complexities. Which of the following is \\textbf{not} possible (We assume that memory access is constant.)?\n\nOptions:\nA. Find the key with precomputation: $0$, memory: $O(1)$, time: $O(N)$.\nB. Find the key with precomputation: $O(N)$, memory: $O(N)$, time: $O(1)$.\nC. Find the key with precomputation: $O(N)$, memory: $O(N^{2/3})$, time: $O(N^{2/3})$.\nD. Find the key with precomputation: $0$, memory: $O(N)$, time: $O(1)$.\n\nAnswer:", | |
| "Question: Given the 2-itemsets {1, 2}, {1, 3}, {1, 5}, {2, 3}, {2, 5}, when generating the 3-itemset we will:?\n\nOptions:\nA. Have 4 3-itemsets after the join and 4 3-itemsets after the prune\nB. Have 4 3-itemsets after the join and 2 3-itemsets after the prune\nC. Have 3 3-itemsets after the join and 3 3-itemsets after the prune\nD. Have 2 3-itemsets after the join and 2 3-itemsets after the prune\n\nAnswer:", | |
| "Question: Which of the following is/are true about stack canaries?\n\nOptions:\nA. The name \u201cstack canaries\u201d stems from birds being used to detect dangerous gas in mines.\nB. Stack canary instrumentation is, by default, inserted into every single function by the compiler.\nC. Stack canaries do not protect against format string vulnerabilities.\nD. Stack canaries on Linux are generated per thread.\n\nAnswer:", | |
| "Question: Does AddressSanitizer prevent \\textbf{all} use-after-free bugs?\n\nOptions:\nA. No, because quarantining free\u2019d memory chunks forever prevents\nB. legit memory reuse and could potentially lead to out-of-memory\nC. No, because UAF detection is not part of ASan's feature set.\nD. situations.\n\nAnswer:", | |
| "Question: Does AddressSanitizer prevent \\textbf{all} use-after-free bugs?\n\nOptions:\nA. Yes, because free\u2019d memory chunks are poisoned.\nB. Yes, because free\u2019d memory is unmapped and accesses therefore\nC. cause segmentation faults.\nD. legit memory reuse and could potentially lead to out-of-memory\n\nAnswer:", | |
| "Question: Consider the list : 5, 12, 7, 8, 41\n\nIn which sorting algorithm(s) do you directly compare 5 and 41 ?\n\nWe are considering the version of insertion sort seen in the slides of the lecture: we compare \\( a_j \\) to \\( a_{j-1} \\), then \\( a_{j-2} \\) etc ...\n\n?\n\nOptions:\nA. Bubble sort\nB. Insertion sort\nC. They are directly compared in both algorithms\nD. They are not compared in any of the algorithms\n\nAnswer:", | |
| "Question: Select the \\emph{incorrect} statement. UMTS uses?\n\nOptions:\nA. MILLENAGE for key establishment.\nB. the block cipher KASUMI.\nC. the block cipher AES.\nD. the block cipher DES.\n\nAnswer:", | |
| "Question: Thick the \\emph{incorrect} assertion.\n\nOptions:\nA. The goal of SAS-based cryptography is to reduce the length of the string that has to be authenticated.\nB. One way to authenticate a SAS is to use your phone.\nC. One can obtain a secure channel from a narrowband authenticated channel using SAS-based cryptography.\nD. SAS-based cryptography always requires the SAS to be collision-resistant.\n\nAnswer:", | |
| "Question: We report the final performance (e.g., accuracy) on the ...\n(One answer)?\n\nOptions:\nA. training \nB. validation \nC. test \nD. all the data together \n\nAnswer:", | |
| "Question: \nYour aim is to evaluate a movie review analysis system, the purpose of which is to determine whether a review is globally positive or negative.\nFor each movie review, such a system outputs one of the following classes: positive and negative.\nTo perform your evaluation, you collect a large set of reviews and have it annotated by two human annotators. This corpus contains 95% of negative reviews (this 95% ratio is for this first question only and may change in the next\n questions).\n\nOptions:\nA. precision\nB. accuracy\nC. Cohen's kappa\nD. recall\n\nAnswer:", | |
| "Question: Which one of the following encryption method is a simple substitution cipher?\n\nOptions:\nA. Vigen\\`ere cipher\nB. the Vernam cipher.\nC. the Caesar cipher.\nD. Enigma\n\nAnswer:", | |
| "Question: Given that $100000000003$ is prime, what is the cardinality of $\\mathbf{Z}_{200000000006}^*$?\n\nOptions:\nA. $2$\nB. $100000000002$\nC. $100000000003$\nD. $200000000006$\n\nAnswer:", | |
| "Question: You are given the following collection of premises:\nIf I go to the museum, it either rains or snows.\nI went to the museum on Saturday or I went to the museum on Sunday.\nIt did not rain and it did not snow on Saturday.\nIt did not rain on Sunday.\nWhich conclusions can be drawn from these premises ? (more than one answer can be correct)?\n\nOptions:\nA. I went to the museum on Sunday.\nB. It snowed on Sunday.\nC. I went to the museum on Saturday.\nD. It was warm on Saturday.\n\nAnswer:", | |
| "Question: Which of the following is \u200btrue \u200bfor a density based cluster C:?\n\nOptions:\nA. Any two points in C must be density reachable. Each point belongs to one, and only one cluster\nB. Any two points in C must be density reachable. Border points may belong to more than one cluster\nC. Any two points in C must be density connected. Border points may belong to more than one cluster\nD. Any two points in C must be density connected. Each point belongs to one, and only one cluster\n\nAnswer:", | |
| "Question: Which of the following statements about libFuzzer is/are correct?\n\nOptions:\nA. if a target can parse several data formats, split it into several\nB. It is better to put narrow targets into the fuzzing stubs, e.g.,\nC. stubs.\nD. targets, one per format.\n\nAnswer:", | |
| "Question: We apply a Gaussian Mixture Model made of $K$ isotropic Gaussians (invariant to rotation around its center) to $N$ vectors of dimension $D$. What is the number of \\emph{free} parameters of this model?\n ?\n\nOptions:\nA. $KD + K - 1$\nB. $NKD$\nC. $NKD + N$\nD. $KD + K + N$\n\nAnswer:", | |
| "Question: Assume in top-1 retrieval recommendation 1 is (2, 3, 1) and recommendation 2 is (2, 1, 3)\n?\n\nOptions:\nA. RMSE(rec 1) < RMSE(rec 2) and DCG(rec 1) > DCG(rec 2)\nB. RMSE(rec 1) = RMSE(rec 2) and DCG(rec 1) > DCG(rec 2)\nC. RMSE(rec 1) < RMSE(rec 2) and DCG(rec 1) = DCG(rec 2)\nD. RMSE(rec 1) = RMSE(rec 2) and DCG(rec 1) = DCG(rec 2)\n\nAnswer:", | |
| "Question: Which of the following is an acceptable commitment scheme, i.e., one that verifies the hiding and binding property (for a well chosen primitive and suitable $x$ and $r$):?\n\nOptions:\nA. $Commit(x;r) = Enc_r(x)$, where $Enc_r$ is a symmetric encryption scheme with key $r$.\nB. $Commit(x;r) = H(x)$, where $H$ is a hash function.\nC. $Commit(x;r) = x \\oplus r$, where $\\oplus$ is the bitwise xor operation.\nD. $Commit(x;r) = H(r\\|x)$, where $H$ is a hash function and $\\|$ denotes the concatenation.\n\nAnswer:", | |
| "Question: Which of the following statements about index merging (when constructing inverted files) is correct?\n\nOptions:\nA. While merging two partial indices on disk, the inverted lists of a term are concatenated without sorting\nB. Index merging is used when the vocabulary does no longer fit into the main memory\nC. The size of the final merged index file is O (n log2 (n) M )), where M is the size of the available memory\nD. While merging two partial indices on disk, the vocabularies are concatenated without sorting\n\nAnswer:", | |
| "Question: Consider a binary classification problem with classifier $f(\\mathbf{x})$ given by $$ f(\\mathbf{x})= \\begin{cases}1, & g(\\mathbf{x}) \\geq 0 \\\\ -1, & g(\\mathbf{x})<0\\end{cases} $$ and $\\mathbf{x} \\in \\mathbb{R}^{6}$. Consider a specific pair $(\\mathbf{x}, y=1)$ and assume that $g(\\mathbf{x})=8$. In particular this means that this point is classified correctly by $f$. Assume further that we have computed the gradient of $g$ at $\\mathbf{x}$ to be $\\nabla_{\\mathbf{x}} g(\\mathbf{x})=(+1,-2,+3,-4,+5,-6)$. You are allowed to make one step in order to (hopefully) find an adversarial example. In the following four questions, assume $\\epsilon=1$. What is the value of $g(\\mathbf{x}+\\delta)$ for this $\\ell_{\\infty}$-optimal choice assuming that $g$ is (locally) linear?\n\nOptions:\nA. $-5$\nB. $+13$\nC. $-7$\nD. $-2$\n\nAnswer:", | |
| "Question: Tick the \\textbf{true} assertion. In a zero-knowledge interactive proof for $L$, \\ldots?\n\nOptions:\nA. for any ppt verifier, there is a simulator which for any $x \\in L$ produces a conversation indistinguishable from the original conversation.\nB. for any ppt verifier, for some $x \\in L$, any simulated conversation is indistinguishable from the original conversation.\nC. the simulator imitates the verifier.\nD. the simulator is computationaly unbounded.\n\nAnswer:", | |
| "Question: If A is an uncountable set and B is an uncountable set, A \u2212 B cannot be :?\n\nOptions:\nA. countably infinite\nB. uncountable\nC. the null set\nD. none of the other options\n\nAnswer:", | |
| "Question: What is the worst case complexity of listing files in a directory? The file system implements directories as hash-tables.\n\nOptions:\nA. $O(number of direntries in the file system)$\nB. $O(1)$\nC. $O(size of the file system)$\nD. $O(number of direntries in the directory)$\n\nAnswer:", | |
| "Question: Is Java as a programming language considered memory safe? Why?\n(note: excluding Java Native Interface, the use of \\texttt{sun.misc.unsafe}, or bugs in the JVM)?\n\nOptions:\nA. Yes, because the JVM guarantees spatial and temporal memory safety at all times.\nB. Yes, because you can't access objects by reference in Java.\nC. No, there are Java programs vulnerable to use after free.\nD. No, there are Java programs vulnerable to type confusion.\n\nAnswer:", | |
| "Question: How do you compute the output Y of a linear regression using Python and the scientific package Numpy? Recall that W is the weight matrix with dimensionality (number of features x 1), b is the bias with dimensionality (1 x 1) and X is the input matrix with dimensionality (number of samples x number of features). Notation of matrix dimensions: (Rows $\\times$ Columns) (One answer)?\n\nOptions:\nA. texttt{Y = X.dot(W) + b}\nB. \\texttt{import numpy as np} \\\\\nC. \\texttt{import numpy as np} \\\\\nD. texttt{Y = X.T.dot(W) + b}\n\nAnswer:", | |
| "Question: Take the recurrence relation $$a_n = -3a_{n-1} + 4a_{n-2}$$ with initial conditions $$a_0 = 1$$, $$a_1=3$$ and transform it in the form $$a_n = \u0007lpha_1r_1^n + \u0007lpha_2r_2^n$$. Which statement is true?\n\nOptions:\nA. $$\u0007lpha_1 = \frac{4}{5}$$ and $$\u0007lpha_2 = \frac{1}{5}$$\nB. $$\u0007lpha_1 = -4$$ and $$\u0007lpha_2=1$$\nC. $$\u0007lpha_1 = \frac{7}{5}$$ and $$\u0007lpha_2 = \frac{-2}{5}$$\nD. $$\u0007lpha_1 = \frac{13}{5}$$ and $$\u0007lpha_2 = \frac{-7}{5}$$\n\nAnswer:", | |
| "Question: Let \\(f(x) = 3 x \\left(\\log_2(x) \right)^2 + x^2 \\). Select the correct statement:?\n\nOptions:\nA. \\(f(x)\\) is \\(O\\left(3 x \\left(\\log_2(x) \right)^2 \right) \\) but not \\(\\Omega\\left(3 x \\left(\\log_2(x) \right)^2 \right) \\)\nB. None of the other statements are correct\nC. \\(f(x)\\) is \\(\\Theta \\left(3 x \\left(\\log_2(x) \right)^2 \right) \\)\nD. \\(f(x)\\) is \\(\\Omega\\left(3 x \\left(\\log_2(x) \right)^2 \right) \\) but not \\(O\\left(3 x \\left(\\log_2(x) \right)^2 \right) \\)\n\nAnswer:", | |
| "Question: Passing on secrets: Alice has posted her RSA credentials as $(m,e)$, with $m$ the modulus and $e$ the encoding exponent. As required by RSA, she keeps her decoding exponent $d$ preciously secret. Bob has a message $t_1$, RSA-encrypts it using $(m,e_1)$ and passes the resulting cryptogram $c_1$ on to Carlos. Carlos has a message $t_2$, RSA-encrypts it using $(m,e_2)$ to obtain the cryptogram $c_2$. Then, Carlos multiplies the two cryptograms, $(c_1\\cdot c_2) \\mod m,$ and passes this to Alice. Alice applies her regular RSA decryption to $(c_1\\cdot c_2) \\mod m.$ Under what condition is the result of this decryption exactly equal to the product $(t_1\\cdot t_2) \\mod m$?\n\nOptions:\nA. If $e_1=e_2=e$.\nB. If $e_1+e_2=e$.\nC. If for some integer $\\ell,$ we have $e_1 e_2 d=\\ell \\phi(m)+1,$ where $\\phi(\\cdot)$ denotes Euler's totient function.\nD. If $d$ is prime and $(e_1+e_2) \\mod m=1$.\n\nAnswer:", | |
| "Question: Your friend Charlie was training a neural net, and observed something very curious. During training, the same network, with the same exact weights, doesn't always give the same prediction for a given example. What can cause this behavior?\n\nOptions:\nA. Adam\nB. SGD with momentum\nC. Dropout\nD. L1 regularization\n\nAnswer:", | |
| "Question: Which of the following statements regarding random forest classifiers is false?\n\nOptions:\nA. Feature bagging has the effect that individual decision trees of a random forest classifier are smaller than a standard decision tree on the same data\nB. Using bootstrapping while training random forests guarantees that all the training data is used during the training phase\nC. Training a random forest classifier can be easily parallelized\nD. Majority voting is an effective way of learning a new classifier based on the predictions of many weak classifiers\n\nAnswer:", | |
| "Question: We consider a month of 30 days. I have \\(n\\) chocolates and each day, I can either: not eat chocolate or eat exactly one chocolate. All chocolates needs to be eaten by the end of the month. What is the smallest number of chocolates needed to guarantee that I will eat chocolate 5 days in a row during the month?\n\nOptions:\nA. 27\nB. 25\nC. 24\nD. 26\n\nAnswer:", | |
| "Question: How many generators do we have in a group of order $13$?\n\nOptions:\nA. 13.\nB. 12.\nC. 6.\nD. 2.\n\nAnswer:", | |
| "Question: You are given a hash function $h$ based on the Merkle-Damg{\\aa}rd scheme. Which of the following attacks is the most difficult, \\emph{a priori}?\n\nOptions:\nA. first preimage attack\nB. collision search\nC. second preimage attack\nD. collision search on the compression function\n\nAnswer:", | |
| "Question: Tick the \\textbf{false} assertion. In Linear Cryptanalysis, the corresponding mask circuit of \\ldots?\n\nOptions:\nA. \\ldots a XOR gate ($X\\oplus Y = Z$) is $a\\cdot Z=(a\\cdot X)\\oplus (a\\cdot Y)$\nB. \\ldots a XOR to constant gate ($Y=X\\oplus K$) is $a\\cdot Y = (a\\cdot X)\\oplus (a\\cdot K)$\nC. \\ldots a linear circuit ($Y=M\\times X$) is $a\\cdot Y = (M\\times a)\\cdot X$\nD. \\ldots a duplicate gate ($X=Y=Z$) is $(a\\oplus b)\\cdot X=(a\\cdot Y)\\oplus (b\\cdot Z)$\n\nAnswer:", | |
| "Question: Consider our standard least-squares problem $$ \\operatorname{argmin}_{\\mathbf{w}} \\mathcal{L}(\\mathbf{w})=\\operatorname{argmin}_{\\mathbf{w}} \\frac{1}{2} \\sum_{n=1}^{N}\\left(y_{n}-\\mathbf{x}_{n}^{\\top} \\mathbf{w}\\right)^{2}+\\frac{\\lambda}{2} \\sum_{d=1}^{D} w_{d}^{2} $$ Here, $\\left\\{\\left(\\mathbf{x}_{n} y_{n}\\right)\\right\\}_{n=1}^{N}$ is the data. The $N$-length vector of outputs is denoted by $\\mathbf{y}$. The $N \\times D$ data matrix is called $\\mathbf{X}$. It's rows contain the tuples $\\mathbf{x}_{n}$. Finally, the parameter vector of length $D$ is called $\\mathbf{w}$. (All just like we defined in the course). Mark any of the following formulas that represent an equivalent way of solving this problem.\n\nOptions:\nA. $\\operatorname{argmin}_{\\mathbf{w}}-\\log p(\\mathbf{y} \\mid \\mathbf{X}, \\mathbf{w}) p(\\mathbf{w})$, where $p(\\mathbf{w})$ correspond to the density of a $D$-length vector of iid zero-mean Gaussians with variance $1 / \\lambda$ and $p(\\mathbf{y} \\mid \\mathbf{X}, \\mathbf{w})$ corresponds to the density of a vector of length $N$ of independent Gaussians of mean $\\mathbf{x}_{n}^{\\top} \\mathbf{w}$, variance 1 and observation $\\mathbf{y}_{n}$ for component $n$.\nB. $\\operatorname{argmin}_{\\mathbf{w}} \\sum_{n=1}^{N}\\left[1-y_{n} \\mathbf{x}_{n}^{\\top} \\mathbf{w}\\right]_{+}+\\frac{\\lambda}{2}\\|\\mathbf{w}\\|^{2}$. Recall: $[z]_{+}=\\max \\{0, z\\}$\nC. $\\operatorname{argmin}_{\\boldsymbol{\\alpha}} \\frac{1}{2} \\boldsymbol{\\alpha}^{\\top}\\left(\\mathbf{X X}^{\\top}+\\lambda \\mathbf{I}_{N}\\right) \\boldsymbol{\\alpha}-\\boldsymbol{\\alpha}^{\\top} \\mathbf{y}$\nD. $\\operatorname{argmin}_{\\mathbf{w}} \\frac{1}{2}\\|\\mathbf{y}-\\mathbf{X} \\mathbf{w}\\|^{2}+\\frac{\\lambda}{2}\\|\\mathbf{w}\\|^{2}$\n\nAnswer:", | |
| "Question: Which of the following statements regarding distant supervision is correct?\n\nOptions:\nA. Distant supervision may suffer from semantic drift\nB. Distant supervision may help to detect complex rules to identify relationships\nC. Distant supervision starts out with hand-written patterns to avoid the use of labeled data\nD. Distant supervision requires manually labeled training data to train a classifier\n\nAnswer:", | |
| "Question: Tick the \\emph{false} assertion. Two-keys triple DES\\dots?\n\nOptions:\nA. is more secure than double encryption.\nB. is less secure than AES.\nC. is as secure as a block cipher using a key twice longer.\nD. is vulnerable to a certain variant of a meet-in-the-middle attacks.\n\nAnswer:", | |
| "Question: Which of the following functions \\( f :\\mathbb{Z} \times \\mathbb{Z} \to \\mathbb{Z} \\) are surjective?\n\nOptions:\nA. \\( f(m,n)=m+n \\)\nB. \\( f(m,n)=m^2+n^2 \\)\nC. \\( f(m,n)=m \\)\nD. \\( f(m,n)=|n| \\)\n\nAnswer:", | |
| "Question: Tick the \\emph{false} assertion. The ElGamal cryptosystem\\dots?\n\nOptions:\nA. is based on the Discrete Logarithm problem.\nB. produces randomized ciphertexts.\nC. produces ciphertexts as long as the plaintexts.\nD. encrypts messages with limited length.\n\nAnswer:", | |
| "Question: When searching for an entity \ud835\udc52\ud835\udc5b\ud835\udc52\ud835\udc64 that has a given relationship \ud835\udc5f with a given entity \ud835\udc52?\n\nOptions:\nA. We search for \ud835\udc52\ud835\udc5b\ud835\udc52\ud835\udc64 that have a similar embedding vector to \ud835\udc52\nB. We search for \ud835\udc52\ud835\udc5b\ud835\udc52\ud835\udc64 that have a similar embedding vector to \ud835\udc52\ud835\udc5c\ud835\udc59\ud835\udc51 which has relationship \ud835\udc5f with \ud835\udc52\nC. We search for pairs (\ud835\udc52\ud835\udc5b\ud835\udc52\ud835\udc64, \ud835\udc52) that have similar embedding to (\ud835\udc52\ud835\udc5c\ud835\udc59\ud835\udc51, \ud835\udc52)\nD. We search for pairs (\ud835\udc52\ud835\udc5b\ud835\udc52\ud835\udc64, \ud835\udc52) that have similar embedding to (\ud835\udc52\ud835\udc5c\ud835\udc59\ud835\udc51, \ud835\udc52) for \ud835\udc52\ud835\udc5c\ud835\udc59\ud835\udc51 which has relationship \ud835\udc5f with \ud835\udc52\n\nAnswer:", | |
| "Question: Consider a binary classification problem with classifier $f(\\mathbf{x})$ given by $$ f(\\mathbf{x})= \\begin{cases}1, & g(\\mathbf{x}) \\geq 0 \\\\ -1, & g(\\mathbf{x})<0\\end{cases} $$ and $\\mathbf{x} \\in \\mathbb{R}^{6}$. Consider a specific pair $(\\mathbf{x}, y=1)$ and assume that $g(\\mathbf{x})=8$. In particular this means that this point is classified correctly by $f$. Assume further that we have computed the gradient of $g$ at $\\mathbf{x}$ to be $\\nabla_{\\mathbf{x}} g(\\mathbf{x})=(+1,-2,+3,-4,+5,-6)$. You are allowed to make one step in order to (hopefully) find an adversarial example. In the following four questions, assume $\\epsilon=1$. Which offset $\\delta$ with $\\|\\delta\\|_{1} \\leq 1$ yields the smallest value for $g(\\mathbf{x}+\\delta)$, assuming that $g$ is (locally) linear?\n\nOptions:\nA. $(+1,-1,+1,-1,+1,-1)$\nB. $(0,0,0,0,0,1)$\nC. $(+1,+1,+1,+1,+1,+1)$\nD. $(+1,-2,+3,-4,+5,-6)$\n\nAnswer:", | |
| "Question: What is $\\lambda (91)$, i.e. the exponent of $\\mathbb{Z}_{91}^*$?\n\nOptions:\nA. 90\nB. 12\nC. 72\nD. 6\n\nAnswer:", | |
| "Question: We consider a classification problem on linearly separable data. Our dataset had an outlier---a point that is very far from the other datapoints in distance (and also far from margins in SVM but still correctly classified by the SVM classifier).\n We trained the SVM, logistic regression and 1-nearest-neighbour models on this dataset.\n We tested trained models on a test set that comes from the same distribution as training set, but doesn't have any outlier points.\n Consider the $K$-fold cross validation on a linear regression model with a sufficiently large amount of training data. When $K$ is large, the computational complexity of the $K$-fold cross validation with respect to $K$ is of order ?\n\nOptions:\nA. $\\mathcal{O}(K)$.\nB. $\\mathcal{O}(K(K-1))$.\nC. $\\mathcal{O}(1)$.\nD. $\\mathcal{O}(1/K)$\n\nAnswer:", | |
| "Question: Which of the following operation is not part of AES?\n\nOptions:\nA. AddRoundKey\nB. SubBytes\nC. MixColumns\nD. MulRoundKey\n\nAnswer:", | |
| "Question: TransE is known to have difficulties in representing symmetric relationships r. Which of the following statements is correct for a symmetric relationship and provides insight for this problem?\n\nOptions:\nA. For all e1 , e2: \\( f(e_1, r, e_2) = f(e_2, r, e_1) \\)\nB. For all e1 , e2: \\( f(e_1, r, e_2) = -f(e_2, r, e_1) \\)\nC. \\( \\Sigma_{ e_1, e_2} f(e_1, r, e_2) + f(e_2, r, e_1) \\)\u00a0is minimized if the embedding vector of r is large\nD. \\( \\Sigma_{ e_1, e_2} f(e_1, r, e_2) + f(e_2, r, e_1) \\) is minimized if the embedding vectors of e1and e2 are close to each other\n\nAnswer:", | |
| "Question: Choose the \\emph{incorrect} statement?\n\nOptions:\nA. The cardinality of an elliptic curve can be bounded using Hasse theorem.\nB. Elliptic curves are defined over a finite field.\nC. Elliptic curves are defined over an Abelian group.\nD. Points on elliptic curves satisfy a cubic polynomial equation.\n\nAnswer:", | |
| "Question: What are the drawbacks of non-preemptive scheduling compared to preemptive scheduling?\n\nOptions:\nA. It can lead to starvation especially for those real-time tasks\nB. Less computational resources need for scheduling and takes shorted time to suspend the running task and switch the context.\nC. Bugs in one process can cause a machine to freeze up\nD. It can lead to poor response time for processes\n\nAnswer:", | |
| "Question: Consider the following joint distribution on $X$ and $Y$, where $X \\in\\{-1,0,1\\}$ and $Y \\in\\{0,1\\}: p(X=-1, Y=0)=0.05$, $p(X=-1, Y=1)=0.05, p(X=0, Y=0)=0.1, p(X=0, Y=1)=0.1, p(X=1, Y=0)=0.3, p(X=1, Y=1)=$ 0.4. You learn that $X \\geq 0$. What is the largest probability of being correct you can achieve when predicting $Y$ in this case?\n\nOptions:\nA. $\\frac{1}{3}$\nB. $\\frac{5}{9}$\nC. $\\frac{1}{2}$\nD. $\\frac{1}{7}$\n\nAnswer:", | |
| "Question: Consider the following set of frequent 3-itemsets: {1, 2, 3}, {1, 2, 4}, {1, 2, 5}, {1, 3, 4}, {2, 3, 4}, {2, 3, 5}, {3, 4, 5}. Which one is not a candidate 4-itemset?\n\nOptions:\nA. {1,2,4,5}\nB. {1,3,4,5} \nC. {2,3,4,5}\nD. {1,2,3,4}\n\nAnswer:", | |
| "Question: Assume we work with $64$-bit block cipher. In differential cryptanalysis, for which of the following circuits is the output difference going to be different from the input difference?\n\nOptions:\nA. a NOT gate\nB. a XOR to a constant gate\nC. a circuit which rotates by $64$ positions to the left\nD. a linear circuit\n\nAnswer:", | |
| "Question: In a Ranked Retrieval result, the result at position k is non-relevant and at k+1 is relevant. Which of the following is always true (P@k and R@k are the precision and recall of the result set consisting of the k top ranked documents)?\n\nOptions:\nA. P@k-1 > P@k+1\nB. P@k-1 = P@k+1\nC. R@k-1 < R@k+\nD. R@k-1 = R@k+1\n\nAnswer:", | |
| "Question: A proof system is perfect-black-box zero-knowledge if \\dots?\n\nOptions:\nA. for any PPT verifier $V$, there exists a PPT simulator $S$, such that $S$ produces an output which is hard to distinguish from the view of the verifier.\nB. for any PPT simulator $S$ and for any PPT verifier $V$, $S^{V}$ produces an output which has the same distribution as the view of the verifier.\nC. there exists a PPT simulator $S$ such that for any PPT verifier $V$, $S^{V}$ produces an output which has the same distribution as the view of the verifier.\nD. there exists a PPT verifier $V$ such that for any PPT simulator $S$, $S$ produces an output which has the same distribution as the view of the verifier.\n\nAnswer:", | |
| "Question: A Carmichael number is?\n\nOptions:\nA. a prime number which cannot pass the Rabin-Miller test.\nB. a composite number which often passes the Rabin-Miller test.\nC. a prime number which cannot pass the Fermat test.\nD. a composite number which often passes the Fermat test.\n\nAnswer:", | |
| "Question: Suppose a file system used only for reading immutable files in random fashion. What is the best block allocation strategy?\n\nOptions:\nA. Linked-list allocation\nB. Continuous allocation\nC. Index allocation with B-tree\nD. Index allocation with Hash-table\n\nAnswer:", | |
| "Question: MD5 is?\n\nOptions:\nA. a secure block cipher\nB. a broken block cipher\nC. a secure hash function\nD. a broken hash function\n\nAnswer:", | |
| "Question: If we have a $n$-bit key, the attack phase with time-memory tradeoff will roughly have\\dots?\n\nOptions:\nA. a time complexity of $1$ (constant).\nB. a time complexity of $2^{\\frac{2}{3}n}$\nC. a time complexity of $2^{\\frac{n}{2}}$\nD. a time complexity of $2^{n}$\n\nAnswer:", | |
| "Question: Consider the document: \u201cInformation retrieval is the task of finding the documents satisfying the information needs of the user\u201d\nUsing MLE to estimate the unigram probability model, what is P(the|Md) and P(information|Md)?\n\nOptions:\nA. 1/16 and 1/16\nB. 1/12 and 1/12\nC. 1/4 and 1/8\nD. 1/3 and 1/6\n\nAnswer:", | |
| "Question: Which of the following are parameters involved in the choice made by an order-1 HMM model for PoS tagging knowing that its output isthis/Pron is/V a/Det good/Adj question/Nand that neither \"is\" nor \"question\" can be adjectives, and that \"question\" can also not be a determiner.(Penalty for wrong ticks.)?\n\nOptions:\nA. P(Pron)\nB. P(Pron is)\nC. P(Pron V)\nD. P(question|N)\n\nAnswer:", | |
| "Question: Alice has been working on a classification problem, and has been using the binary cross-entropy loss function, defined as: $\\mathcal{L}_{\\text{BCE}}(\\mathbf{y}, \\mathbf{\\hat{y}})=- \\frac{1}{N}\\sum^{N}_{i=1} y^{(i)} \\log(\\hat{y}^{(i)}) + (1-y^{(i)}) \\log(1- \\hat{y}^{(i)})$.\\\\\n\n Despite trying many models, she hasn't been able to reduce the training loss. Her friend Frank suggests using a new loss function he invented, which he proudly calls the \"Frank Loss\", defined as: $\\mathcal{L}_{\\text{Frank}}(\\mathbf{y}, \\mathbf{\\hat{y}})= -e^{-1} + \\mathcal{L}_{\\text{BCE}}(\\mathbf{y}, \\mathbf{\\hat{y}})$.\n After switching to the Frank loss, Alice notices that the training loss is immediately lower! How will that affect the training accuracy?\n\nOptions:\nA. The training accuracy will increase.\nB. The training accuracy will decrease.\nC. The training accuracy will stay the same.\nD. It is impossible to say without more information.\n\nAnswer:", | |
| "Question: Confidentiality means that:?\n\nOptions:\nA. the message can be read by anyone.\nB. information should not leak to any unexpected party.\nC. the message should make clear who the author is.\nD. the information must be protected against any malicious modification.\n\nAnswer:", | |
| "Question: In linear cryptanalysis,\\dots?\n\nOptions:\nA. one needs to do a chosen plaintext attack.\nB. one studies how the differences in the input propagate in the cipher.\nC. one chooses the deviant property with the smallest bias in order to optimize the attack.\nD. one needs to have about $\\frac{1}{LP}$ pairs of plaintext-ciphertext in order to recover the correct key, where $LP$ is the linear probability of the cipher.\n\nAnswer:", | |
| "Question: Consider the following CF grammar \\(G_1\\)\n\nOptions:\nA. (0.55, 0.45, 0.60, 0.10, 0.15, 0.75, 0.50, 0.50)\nB. (1.00, 0.00, 1.00, 0.00, 1.00, 0.00, 1.00, 0.00)\nC. I cannot answer because it also depends on the probabilistic coefficients associated to the lexical rules.\nD. (0.35, 0.65, 1.00, 0.30, 0.25, 0.45, 0.25, 0.75)\n\nAnswer:", | |
| "Question: Let $s$ be a security parameter and $n$ be a constant. Which of the following functions is negligible?\n\nOptions:\nA. $1/2$.\nB. $1/s$.\nC. $1/e^s$.\nD. $1/s^n$.\n\nAnswer:", | |
| "Question: Determine which of the following compound propositions are satisfiable (more than one answer can be correct):?\n\nOptions:\nA. (p\u2228\u00acq)\u2227(\u00acp\u2228q)\u2227(\u00acp\u2228\u00acq)\nB. (p\u2194q)\u2227(\u00acp\u2194q)\nC. (p \u2192 q)\u2227(p \u2192 \u00acq)\u2227(\u00acp \u2192 q)\nD. None of the other options\n\nAnswer:", | |
| "Question: For his awesome research, Tugrulcan is going to use the Pagerank with teleportation and HITS algorithm, not on a network of webpages but on the retweet network of Twitter! The retweet network is a directed graph, where nodes are users and an edge going out from a user A and to a user B means that \"User A retweeted User B\". Which one is FALSE about a Twitter bot that retweeted other users frequently but got never retweeted by other users or by itself?\n\nOptions:\nA. It will have a non-zero hub value.\nB. It will have an authority value of zero.\nC. It will have a pagerank of zero.\nD. Its authority value will be equal to the hub value of a user who never retweets other users.\n\nAnswer:", | |
| "Question: Tick the \\textbf{true} statement.\n\nOptions:\nA. The Legendre Symbol is the generalization of the Jacobi Symbol.\nB. The complexity to compute $\\left ( \\frac{a}{n} \\right )$ is $\\mathcal{O}(\\ell^2)$, where $\\ell$ is the length of $a$ and $n$.\nC. If $\\left ( \\frac{a}{n} \\right )=1$, then $a$ is a quadratic residue in $\\mathbb{Z}_n^*$.\nD. For all $a$ and $b$ ($b$ odd) $\\left ( \\frac{a}{b} \\right )=0$ if and only if $b$ divides $a$.\n\nAnswer:", | |
| "Question: Which flag prevents user programs from reading and writing kernel data?\n\nOptions:\nA. PTE_P\nB. PTE_U\nC. PTE_D\nD. PTE_W\n\nAnswer:", | |
| "Question: Let $(G,+)$ be a group of order $n$. If $g$ is a generator of this group, then it has order\\dots?\n\nOptions:\nA. $n/2$\nB. $\\sqrt{n}$\nC. $n-1$\nD. $n$\n\nAnswer:", | |
| "Question: Consider a classification problem on linearly separable data. We train an SVM model and a logistic regression model. For logistic regression (LR) we add a small regularization term (penalty on weights) in order to make the optimum well-defined. Each model gives us a margin. Consider a datapoint $\\mathbf{x}_{0}$ that is correctly classified and strictly outside both margins Which one of the following statements is incorrect ?\n\nOptions:\nA. There exists a direction in which we can arbitrarily move $\\mathbf{x}_{0}$ without changing the SVM decision boundary after retraining.\nB. $\\mathbf{x}_{0}$ isn't a support vector\nC. There exists a direction in which we can slightly move $\\mathbf{x}_{0}$ without changing the LR decision boundary after retraining.\nD. If we remove $\\mathbf{x}_{0}$ from the dataset and retrain, this will change the LR decision boundary.\n\nAnswer:", | |
| "Question: Consider 3 regular expressions \\(A\\), \\(B\\), and \\(C\\), such that:the sets of strings recognized by each of the regular expressions is non empty;the set of strings recognized by \\(B\\) is included in the set of strings recognized by \\(A\\);some strings are recognized simultaneously by \\(A\\) and by \\(C\\); andno string is recognized simultaneously by \\(B\\) and \\(C\\).Which of the following statements are true?(where, for a regular expression \\(X\\),\u00a0 \\((X)\\) denotes the transducer which associates every string recognized by \\(X\\) to itself)(Penalty for wrong ticks)?\n\nOptions:\nA. Any string recognized by \\(A\\) but not by \\(B\\) is a left string in an association recognized by the transducer \\((A)(C)\\)\nB. Any string recognized by \\(B\\) is (at least) associated to itself by the transducer \\(A\\otimes B\\)\nC. \\((A\\otimes B)\\circ (C)\\) recognizes a non empty set of string associations\nD. \\((B\\otimes A)\\circ (C)\\) recognizes a non empty set of string associations\n\nAnswer:", | |
| "Question: For this question, one or more assertions can be correct. Tick only the correct assertion(s). There\nwill be a penalty for wrong assertions ticked.Using a 3-gram character model, which of the following expressions are equal to \\( P(\\text{opossum}) \\) ?\n\nOptions:\nA. \\( P(\\text{opo}) \\cdot P(\\text{pos}) \\cdot P(\\text{oss}) \\cdot P(\\text{ssu}) \\cdot P(\\text{sum}) \\)\nB. \\( P(\\text{op}) \\cdot P(\\text{op}|\\text{o}) \\cdot P(\\text{po}|\\text{s}) \\cdot P(\\text{os}|\\text{s}) \\cdot P(\\text{ss}|\\text{u}) \\cdot P(\\text{su}|\\text{m}) \\)\nC. \\( P(\\text{op}) \\cdot P(\\text{o}|\\text{po}) \\cdot P(\\text{s}|\\text{op}) \\cdot P(\\text{s}|\\text{so}) \\cdot P(\\text{u}|\\text{ss}) \\cdot P(\\text{m}|\\text{us}) \\)\nD. \\( P(\\text{op}) \\cdot P(\\text{o}|\\text{op}) \\cdot P(\\text{s}|\\text{po}) \\cdot P(\\text{s}|\\text{os}) \\cdot P(\\text{u}|\\text{ss}) \\cdot P(\\text{m}|\\text{su}) \\)\n\nAnswer:", | |
| "Question: KEM/DEM refers to\\dots?\n\nOptions:\nA. an encryption scheme.\nB. a digital signature scheme.\nC. a commitment scheme.\nD. a hash function.\n\nAnswer:", | |
| "Question: Which of the following statements about fuzzing is/are correct?\n\nOptions:\nA. Greybox fuzzing keeps track of concrete program paths to\nB. Greybox fuzzing is always the better alternative to\nC. Generational fuzzing requires more manual work (to specify the\nD. blackbox fuzzing.\n\nAnswer:", | |
| "Question: Which of the following hold(s) true about update deployment in the secure\n development lifecycle?\n\nOptions:\nA. ensure that users have the latest code installed.\nB. Updates may bring new code that may be buggy, so additional\nC. monitoring is required after deploying an update.\nD. One motivation for automatic updates is for manufacturers to\n\nAnswer:", | |
| "Question: In principal component analysis, the left singular vectors $\\mathbf{U}$ of a data matrix $\\mathbf{X}$ of shape ( $d$ features, $n$ datapoints) are used to create a new data matrix $\\mathbf{X}^{\\prime}=\\mathbf{U}^{\\top} \\mathbf{X}$. To achieve dimensionality reduction, we keep only certain rows of the matrix $\\mathbf{X}^{\\prime}$. We keep those rows that have:?\n\nOptions:\nA. the lowest variance.\nB. the highest variance.\nC. smallest L2 norm.\nD. L2 norm closest to 1\n\nAnswer:", | |
| "Question: Tick the \\emph{incorrect} statement for \\emph{independent} random variables $X$ and $Y$.\n\nOptions:\nA. $Var(XY) = Var(X)Var(Y)$.\nB. $E[XY] = E[X]E[Y]$.\nC. $\\Pr[X=x\\text{ and } Y = y] = \\Pr[X=x]\\Pr[Y=y]$.\nD. $E[X+Y] = E[X] + E[Y]$.\n\nAnswer:", | |
| "Question: The Diffie-Hellman key agreement protocol \\ldots?\n\nOptions:\nA. \\ldots was invented by Rivest, Shamir and Adleman.\nB. \\ldots can be broken with a factoring oracle.\nC. \\ldots resists to active adversaries.\nD. \\ldots resists to passive adversaries.\n\nAnswer:", | |
| "Question: Tick the \\emph{wrong} assertion concerning 3G.\n\nOptions:\nA. In 3G, the network is authenticated to the phone.\nB. The integrity of 3G messages is well protected.\nC. In 3G, there is a counter to protect against replay attacks.\nD. 3G uses f8 for encryption.\n\nAnswer:", | |
| "Question: In an FP tree, the leaf nodes are the ones with:?\n\nOptions:\nA. Lowest confidence\nB. Lowest support\nC. Least in the alphabetical order\nD. None of the other options.\n\nAnswer:", | |
| "Question: In JOS, suppose a value is passed between two Envs. What is the minimum number of executed system calls?\n\nOptions:\nA. 1\nB. 2\nC. 3\nD. 4\n\nAnswer:", | |
| "Question: Which of the following graph analysis techniques do you believe would be most appropriate to identify communities on a social graph?\n\nOptions:\nA. Cliques\nB. Random Walks\nC. Shortest Paths\nD. Association rules\n\nAnswer:", | |
| "Question: Select \\emph{incorrect} statement. Generic attacks on DES include?\n\nOptions:\nA. time memory tradeof against 2 key Triple DES.\nB. collision attack against 3 key Triple DES.\nC. meet in the middle attack against 3 key Triple DES.\nD. known plaintext attack by Van Oorschot-Wiener agains 2 key Triple DES.\n\nAnswer:", | |
| "Question: Tick the \\textbf{correct} assertion.\n\nOptions:\nA. Forward secrecy means that an email should still be encrypted when we forward it to someone else.\nB. Forward secrecy means that the communication will remain secure even when the long term secrets keys are disclosed.\nC. Semi-static Diffie-Hellman provides forward secrecy.\nD. Static Diffie-Hellman provides forward secrecy.\n\nAnswer:", | |
| "Question: Let P(x) is \u201cx is an elephant\u201d and F(x) is \u201cx flies\u201d and the domain consists of all animals. Translate the following statement into English: \u2203!x(P(x) \u2227 F(x))?\n\nOptions:\nA. All elephants fly\nB. There is an animal that flies if it\u2019s an elephant\nC. There exists only one elephant that can fly\nD. Some elephants can flyv\n\nAnswer:", | |
| "Question: In order to have security of $2^{128}$ against key recovery attack for a (secure) symmetric key encryption scheme \\ldots?\n\nOptions:\nA. we should use DES!\nB. the shortest secret key that we can use is $128$ bits long.\nC. the shortest secret key that we can use is $64$ bits long.\nD. the longest secret key that we can use is $256$ bits long.\n\nAnswer:", | |
| "Question: A page that points to all other pages but is not pointed by any other page would have...\n\nOptions:\nA. Zero hub\nB. Nonzero pagerank\nC. Nonzero authority\nD. None of the above\n\nAnswer:", | |
| "Question: Let $N=3\\cdot5\\cdot13$. Then a quadratic residue in $\\mathbb{Z}_N^*$ has?\n\nOptions:\nA. $2$ square roots.\nB. $3$ square roots.\nC. $6$ square roots.\nD. $8$ square roots.\n\nAnswer:", | |
| "Question: We represent $GF(2^8)$ as $\\mathbb{Z}_2[X]/P(X)$ where $P(X) = X^8 + X^4+X^3+X+1$. Then, $(X^7+X^6)\\times (X + 1)=$\\dots?\n\nOptions:\nA. $X^6+X^5+X^4+X^3+X$.\nB. $X^6 + X^4 + X^3 + X + 1$.\nC. $X^6$.\nD. $X^7+X^6+X^4+X^3+X+1$.\n\nAnswer:", | |
| "Question: Consider a binary classification problem with classifier $f(\\mathbf{x})$ given by $$ f(\\mathbf{x})= \\begin{cases}1, & g(\\mathbf{x}) \\geq 0 \\\\ -1, & g(\\mathbf{x})<0\\end{cases} $$ and $\\mathbf{x} \\in \\mathbb{R}^{6}$. Consider a specific pair $(\\mathbf{x}, y=1)$ and assume that $g(\\mathbf{x})=8$. In particular this means that this point is classified correctly by $f$. Assume further that we have computed the gradient of $g$ at $\\mathbf{x}$ to be $\\nabla_{\\mathbf{x}} g(\\mathbf{x})=(+1,-2,+3,-4,+5,-6)$. You are allowed to make one step in order to (hopefully) find an adversarial example. In the following four questions, assume $\\epsilon=1$. Which offset $\\delta$ with $\\|\\delta\\|_{\\infty} \\leq 1$ yields the smallest value for $g(\\mathbf{x}+\\delta)$, assuming that $g$ is (locally) linear?\n\nOptions:\nA. $(+1,-2,+3,-4,+5,-6)$\nB. $-(0,0,0,0,0,1)$\nC. $(0,0,0,0,0,1)$\nD. $(-1,+2,-3,+4,-5,+6)$\n\nAnswer:", | |
| "Question: For each of the sub-questions of this question (next page), tick/check the corresponding box if the presented sentence is correct\nat the corresponding level (for a human). There will be a penalty for wrong boxes ticked/checked.The Times is a British daily national\nnewspaper based in London.\n\nOptions:\nA. syntactic\nB. pragmatic\nC. lexical\nD. semantic\n\nAnswer:", | |
| "Question: The feature ``deck structure type'' can have the following values: \n Cast-in-place Concrete,\n Concrete Precast Panel,\n Open Grating,\n Closed Grating,\n Steel plate,\n Corrugated Steel,\n Aluminum and\n Timber.\n\tFor logistic regression, what is the best encoding for these values? (One or multiple answers)?\n\nOptions:\nA. assign an integer to each option\nB. one-hot encoding\nC. polynomial encoding\nD. logistic regression does not require an encoding\n\nAnswer:", | |
| "Question: If for the \u03c72 statistics for a binary feature we obtain P(\u03c72 | DF = 1) < 0.05 this means?\n\nOptions:\nA. That the two features are correlated.\nB. That the class label is independent of the feature\nC. That the class label correlates with the feature\nD. None of the above\n\nAnswer:", | |
| "Question: Assume Alice and Bob are exchanging messages that are encrypted with the Vernam cipher. Their communication therefore necessarily has \\dots?\n\nOptions:\nA. confidentiality only.\nB. confidentiality and integrity.\nC. confidentiality, integrity and authenticity.\nD. none of the mentioned properties.\n\nAnswer:", | |
| "Question: Suppose there is a room with $n$ people and you want to generate a secret key for all pairs of participants. You need?\n\nOptions:\nA. $2^n$ keys.\nB. $n^2$ keys.\nC. $\\log_2(n-1)$ keys.\nD. $\\frac{n(n-1)}{2}$ keys.\n\nAnswer:", | |
| "Question: You are doing your ML project. It is a regression task under a square loss. Your neighbor uses linear regression and least squares. You are smarter. You are using a neural net with 10 layers and activations functions $f(x)=3 x$. You have a powerful laptop but not a supercomputer. You are betting your neighbor a beer at Satellite who will have a substantially better scores. However, at the end it will essentially be a tie, so we decide to have two beers and both pay. What is the reason for the outcome of this bet?\n\nOptions:\nA. Because we use exactly the same scheme.\nB. Because it is almost impossible to train a network with 10 layers without a supercomputer.\nC. Because I should have used more layers.\nD. Because I should have used only one layer.\n\nAnswer:", | |
| "Question: Tick the \\emph{correct} assertion. The maximum advantage of an \\textbf{adaptive} distinguisher limited to $q$ queries between two random functions $F$ and $F^*$ is always\\dots?\n\nOptions:\nA. $\\frac{1}{2}|||[F]^q - [F^*]^q |||_{\\infty}$.\nB. $\\frac{1}{2}|||[F]^q - [F^*]^q |||_{a}$.\nC. $1$ when $F = F^*$.\nD. lower than the advantage of the best \\textbf{non-adaptive} distinguisher.\n\nAnswer:", | |
| "Question: How do you compute the output Y of a linear regression using Python and the scientific package Numpy? Recall that W is the weight matrix with dimensionality (number of features x 1), b is the bias with dimensionality (1 x 1) and X is the input matrix with dimensionality (number of samples x number of features). Notation of matrix dimensions: (Rows $\\times$ Columns) (One answer)?\n\nOptions:\nA. \\texttt{import numpy as np} \\\\\nB. \\texttt{import numpy as np} \\\\\nC. texttt{Y = X.dot(W) + b}\nD. texttt{Y = W.dot(X) + b}\n\nAnswer:", | |
| "Question: Which of the following in Linux x86-64 assembly snippets can be used as a\n gadget AND can be chained with more gadgets (e.g., in a ROP/JOP chain)?\n\nOptions:\nA. \\texttt{pop rbx; pop rax; jmp rax}\nB. \\texttt{pop rbx; pop rax; ret}\nC. \\texttt{mov eax, -1; call rax}\nD. \\texttt{xor rbx, rbx; xor rbx, -1; push rbx; ret}\n\nAnswer:", | |
| "Question: Which of the following statements is correct in the context of \u00a0information extraction?\n\nOptions:\nA. A confidence measure that prunes too permissive patterns discovered with bootstrapping can help reducing semantic drift\nB. The bootstrapping technique requires a dataset where statements are labelled\nC. Distant supervision typically uses low-complexity features only, due to the lack of training data\nD. For supervised learning, sentences in which NER has detected no entities are used as negative samples\n\nAnswer:", | |
| "Question: How is data integrity ensured in WEP?\n\nOptions:\nA. there is no integrity.\nB. A3.\nC. Michael.\nD. HMAC.\n\nAnswer:", | |
| "Question: For any function $f:\\{0,1\\}^p\\rightarrow \\{0,1\\}^q$ and for any $a\\in\\{0,1\\}^p$, we have\\ldots?\n\nOptions:\nA. $\\Sigma _{b\\in \\{0,1\\}^q}\\mathsf{DP}^f(a,b)=1$\nB. $\\Sigma _{b\\in \\{0,1\\}^q}\\mathsf{DP}^f(a,b)=0$\nC. $\\Sigma _{b\\in \\{0,1\\}^q}\\mathsf{DP}^f(a,b)=\\frac{1}{2}$\nD. $\\Sigma _{b\\in \\{0,1\\}^q}\\mathsf{DP}^f(a,b)=\\frac{1}{\\sqrt{2}}$\n\nAnswer:", | |
| "Question: Which of the following is well preserved by 2G?\n\nOptions:\nA. Confidentiality\nB. Message Integrity\nC. Challenge freshness\nD. Authentication of Mobile Station\n\nAnswer:", | |
| "Question: Which of the following mitigation(s) can prevent reliable exploitation of\n the following vulnerable code snippet?\n\nOptions:\nA. Fortify source\nB. ASLR for position-independent executables\nC. DEP\nD. Stack canaries\n\nAnswer:", | |
| "Question: Which loss function(s) should you use? (One or multiple answers)?\n\nOptions:\nA. mean square error (MSE) loss\nB. cross entropy loss\nC. hinge loss\nD. L1 loss\n\nAnswer:", | |
| "Question: You want to build a convolutional neural network to distinguish between types of cars in images. Your friend Alice, a biologist, has been working on a network to classify wildlife, which she calls WildNet. She spent several weeks training that network, and made it accessible to you. What can you do with it?\n\nOptions:\nA. I can't reuse any of the weights of any layer, but I can take inspiration from the architecture of WildNet.\nB. Nothing, wildlife is not the same as cars.\nC. I can freeze the early layers, and then remove the last layers and replace it with my own re-learned layers. That way, I can make use of the generic features learned by WildNet.\nD. I can freeze the last few layers, and then remove the early layers and replace it with my own re-learned layers. That way, I can make use of the generic features learned by WildNet.\n\nAnswer:", | |
| "Question: Select the \\emph{correct} statement. In DSA signature, i.e., DSS?\n\nOptions:\nA. the public parameter is $N = pq$, for $p$ and $q$ primes, such that $p=aq + 1$.\nB. public parameters are primes $p$ and $q$ such that $p=aq + 1$.\nC. the public parameter is a random number $N$ such that $\\gcd(N, T) = 1$, where $T$ is defined by the DSS standard.\nD. does not require any public parameter.\n\nAnswer:", | |
| "Question: Tick the \\emph{false} assertion about Diffie and Hellman.\n\nOptions:\nA. They wrote an article entitled ``\\emph{New directions in Cryptography}'' in 1976.\nB. They introduced the notion of ``\\emph{trapdoor permutation}''.\nC. They proposed a key agreement protocol.\nD. They invented RSA.\n\nAnswer:", | |
| "Question: Tick the \\textbf{true} assertion.\n\nOptions:\nA. Luby-Rackoff's lemma bounds the advantage of distinguisher trying to discern a 3-round Feistel scheme from another Feistel scheme.\nB. Decorrelation can express the best $d$-limited non-adaptive distinguisher between two random functions $F$ and $G$.\nC. Decorrelation uses the $a$-norm to express the advantage gained by the best, limited non-adaptive distinguisher.\nD. Decorrelation is in fact concerned only with expressing the advantage of adaptive distinguishers.\n\nAnswer:", | |
| "Question: The exponent of the group $\\mathbb{Z}_9^*$ is?\n\nOptions:\nA. 6.\nB. 9.\nC. 8.\nD. 3.\n\nAnswer:", | |
| "Question: In Machine Learning, we want to learn the \\textbf{parameters W} for the mapping function f: $y=f(x,W) +\\epsilon$ where x is the input, y the output, and $\\epsilon$ the error term.\\\\\n(One or multiple answers)?\n\nOptions:\nA. When f: $R \\rightarrow \\{1,..N\\}$, it is a classification task\nB. When f: $R^M \\rightarrow \\{1,..N\\}$, it is a classification task\nC. When f: $R^M \\rightarrow R$, it is a classification task \nD. When f: $R^M \\rightarrow R$, it is a regression task\n\nAnswer:", | |
| "Question: What should the minimal length of the output of a hash function be to provide security against \\emph{collision attacks} of $2^{256}?$?\n\nOptions:\nA. $2^{256}$ bits.\nB. $2^{512}$ bits.\nC. $256$ bits.\nD. $512$ bits.\n\nAnswer:", | |
| "Question: Consider the plain Rabin cryptosystem with modulus $n=pq$ with two large primes $p,q$. Tick the \\textit{incorrect} assertion.\n\nOptions:\nA. The encryption algorithm has lower complexity than the decryption algorithm.\nB. The decryption problem is equivalent to the problem of computing square roots in $\\mathbb{Z}_n^*$.\nC. Without using any redundancy in the messages, for every ciphertexts we can have up to 3 possible plaintexts.\nD. The key recovery problem is equivalent to the problem of factoring $n$.\n\nAnswer:", | |
| "Question: Let $h$ be a cryptographic hash function based on the Merkle-Damg{\\aa}rd scheme. The Merkle-Damg{\\aa}rd Theorem states that\\dots?\n\nOptions:\nA. \\dots $h$ is collision-resistant.\nB. \\dots $h$ is resistant to a first preimage attack.\nC. \\dots if the compression function is collision-resistant, then $h$ is collision-resistant.\nD. \\dots if $h$ is collision-resistant, then the compression function is collision-resistant.\n\nAnswer:", | |
| "Question: Given the distribution of positive and negative samples for attributes A1 and A2, which is the best attribute for splitting?\n\n|A1|P|N|\n|---|---|---|\n|a|7|0|\n|b|1|4|\n\n|A2|P|N|\n|--|--|--|\n|x|5|1|\n|y|3|3|?\n\nOptions:\nA. A1\nB. A2\nC. They are the same\nD. There is not enough information to answer the question\n\nAnswer:", | |
| "Question: Which of the following algorithms is \\emph{not} a hash function?\n\nOptions:\nA. SHA-1\nB. MD5\nC. RC4\nD. MD4\n\nAnswer:", | |
| "Question: We apply a Gaussian Mixture Model made of $K$ isotropic Gaussians (invariant to rotation around its center) to $N$ vectors of dimension $D$. What is the number of \\emph{free} parameters of this model?\n ?\n\nOptions:\nA. $KD + K - 1$\nB. $NKD + NKD^2$\nC. $2KD + N - 1$\nD. $2KD - 1$\n\nAnswer:", | |
| "Question: In Text Representation learning, which of the following statements are correct?\n\nOptions:\nA. Learning GloVe word vectors can be done using the singular value decomposition, if the $f_{d n}$ weights are set to 1 for all observed entries.\nB. The skip-gram model for learning original word2vec embeddings does learn a binary classifier for each word.\nC. FastText as discussed in the course learns word vectors and sentence representations which are specific to a supervised classification task.\nD. Logistic regression used for text classification is faster at test time when using word vectors as opposed to bag-of-word representation of the input.\n\nAnswer:", | |
| "Question: Consider a movie recommendation system which minimizes the following objective\n \frac{1}{2} \\sum_{(d,n)\\in\\Omega} [x_{dn} - (\\mathbf{W} \\mathbf{Z}^\top)_{dn}]^2 + \frac{\\lambda_w}{2} \norm{\\mathbf{W}}_\text{Frob}^2 + \frac{\\lambda_z}{2} \norm{\\mathbf{Z}}_\text{Frob}^2\n where $\\mathbf{W}\\in \\R^{D\times K}$ and $\\mathbf{Z}\\in \\R^{N\times K}$.\n\t Suppose movies are divided into genre A and genre B (i.e., $\\mathbf{W}_A\\in \\R^{D_A\times K}, \\mathbf{W}_B\\in \\R^{D_B\times K}, \\mathbf{W}=[\\mathbf{W}_A; \\mathbf{W}_B]$, with $D_A\\!+\\!D_B=D$) and users are divided into group 1 and group 2 (i.e., $\\mathbf{Z}_1\\in \\R^{N_1\times K}, \\mathbf{Z}_2\\in \\R^{N_2\times K}, \\mathbf{Z}=[\\mathbf{Z}_1; \\mathbf{Z}_2]$, with $N_1\\!+\\!N_2=N$). In addition, group 1 users only rate genre A movies while group 2 users only rate genre B movies. Then instead of training a large recommendation system with $(\\mathbf{W}, \\mathbf{Z})$, one may train two smaller recommendation systems with parameters $(\\mathbf{W_A}, \\mathbf{Z_1})$ and $(\\mathbf{W_B}, \\mathbf{Z_2})$ separately. If SGD is used to solve the minimization problems and all conditions remain the same (e.g., hyperparameters, sampling order, initialization, etc), then which of the following statements is true about the two training methods?\n\nOptions:\nA. Feature vectors obtained in both cases remain the same. \nB. Feature vectors obtained in both cases are different.\nC. Feature vectors obtained in both cases can be either same or different, depending on the sparsity of rating matrix.\nD. Feature vectors obtained in both cases can be either same or different, depending on if ratings in two groups and genres are evenly distributed.\n\nAnswer:", | |
| "Question: We want to return, from the two posting lists below, the top-2 documents matching a\nquery using Fagin\u2019s algorithm with the aggregation function taken as the sum of the tf-idf weights. How many entries (total of both lists) are accessed in the first phase of the\nalgorithm performing round-robin starting at List 1 (i.e., before performing the random\naccess)?\n\nOptions:\nA. 4\nB. 6\nC. 8\nD. 10\n\nAnswer:", | |
| "Question: Tick the \\emph{incorrect} assertion. A cipher $C$ perfectly decorrelated at order 2 implies\\dots?\n\nOptions:\nA. perfect secrecy when used twice.\nB. security against differential cryptanalysis.\nC. security against linear cryptanalysis.\nD. security against exhaustive search.\n\nAnswer:", | |
| "Question: Tick the \\textit{incorrect} assertion.\n\nOptions:\nA. Plain CBCMAC resists forgery attacks.\nB. GCM is a blockcipher mode of operation that provides both confidentiality and authenticity for messages.\nC. A message authentication scheme that resists a chosen message forgery attack will also resist a known message forgery attack.\nD. HMAC is a message authentication code based on a hash function.\n\nAnswer:", | |
| "Question: Which of the following integers has the square roots $\\{2,3\\}$ when taken modulo $5$ \\textbf{and} the square roots $\\{3,10\\}$ when taken modulo $13$.\n\nOptions:\nA. $4$.\nB. $9$.\nC. $6$.\nD. $5$.\n\nAnswer:", | |
| "Question: Given a document collection, if we change the ordering of the words in the documents, which of the following will not change?\n\nOptions:\nA. Singular values in Latent Semantic Indexing (LSI)\nB. The entities extracted using a Hidden Markov Model (HMM)\nC. The embedding vectors produced by Word2vec\nD. All the previous will change\n\nAnswer:", | |
| "Question: According to the Kerckhoffs Principle:?\n\nOptions:\nA. The internal design of a cryptosystem should be public.\nB. The internal design of a cryptosystem should \\emph{not} be public.\nC. If there is a single security hole in a cryptosystem, somebody will discover it.\nD. The security of the cryptosystem should \\emph{not} rely on the secrecy of the cryptosystem itself.\n\nAnswer:", | |
| "Question: Consider a neural net with $K$ nodes per hidden layer. In a standard such net we have $K^{2}$ parameters/weights (ignoring the bias terms) per layer. Consider a convolutional net where the data is laid out in a one-dimensional fashion and the filter/kernel has $M$ non-zero terms. Ignoring the bias terms, how many parameters are there per layer?\n\nOptions:\nA. (b) $M^{2}$\nB. (d) $K$\nC. (a) $K^{2}$\nD. (c) $K M$\n\nAnswer:", | |
| "Question: In a zero-knowledge interactive proof, the prover wants to prove :?\n\nOptions:\nA. nothing.\nB. a statement without interaction.\nC. something he doesn't know.\nD. a statement without revealing anything else besides that the statement holds.\n\nAnswer:", | |
| "Question: The Merkle-D{\\aa}mgard construction is?\n\nOptions:\nA. a method which iterates a hash function to obtain a compression function.\nB. a method which iterates a compression function to obtain a hash function.\nC. a method which constructs a compression function from a block cipher.\nD. a method which constructs a block cipher function from a hash function.\n\nAnswer:", | |
| "Question: Given this program snippet which is part of a large (> 10000 LoC) codebase,\n which of these statements are true, given that the contents of string \"s\"\n are attacker controlled, the attacker can run the function f only once, the\n attacker has access to the binary and the binary is compiled for x86\\_64 on\n a modern Linux system?\n\nOptions:\nA. mitigation, an attacker can leak the canary.\nB. and no other mitigation, an attacker can gain remote code execution.\nC. If this program is compiled with stack canaries and no other\nD. If this program is compiled with stack canaries and no other\n\nAnswer:", | |
| "Question: For two $\\ell$-bit numbers, the \\textbf{tightest} complexity to compute their gcd is?\n\nOptions:\nA. $\\mathcal{O}(2^{\\ell})$.\nB. $\\mathcal{O}(\\ell^3)$.\nC. $\\mathcal{O}(\\ell^2)$.\nD. $\\mathcal{O}(\\log{\\ell}^2)$.\n\nAnswer:", | |
| "Question: Which of the following statements about index merging (when constructing inverted files) is correct?\n\nOptions:\nA. While merging two partial indices on disk, the inverted lists of a term are concatenated without sorting\nB. Index merging is used when the vocabulary does no longer fit into the main memory\nC. The size of the final merged index file is O(nlog2(n)*M), where M is the size of the available memory\nD. While merging two partial indices on disk, the vocabularies are concatenated without sorting\n\nAnswer:", | |
| "Question: You need to debug your Stochastic Gradient Descent update for a classification of three bridge types.\n\tManually compute the model output for the feature vector $x=(1, 0, 0, 0, 0)$ and $W$ contains only zeros. The model is logistic regression, \\textit{i.e.}, $\\textrm{softmax}(Wx)$. \n\tRemember:\n\t\\begin{equation}\n\t \\textrm{softmax}_i(s) = \\frac{e^{s_i}}{\\sum_k e^{s_k}}\n\t\\end{equation}\n\t(One answer!!!!!!)?\n\nOptions:\nA. $(0, 0, 0)$\nB. $(\\frac{1}{3}, \\frac{1}{3}, \\frac{1}{3})$\nC. $(0, 0, 0, 0, 0)$\nD. $(\\frac{1}{5}, \\frac{1}{5}, \\frac{1}{5}, \\frac{1}{5}, \\frac{1}{5})$\n\nAnswer:", | |
| "Question: Consider the group $(\\mathbb{Z} / 23 \\mathbb{Z}^*, \\cdot)$. Find how many elements of the group are generators of the group. (Hint: $5$ is a generator of the group.)?\n\nOptions:\nA. $10$\nB. $22$\nC. $11$\nD. $2$\n\nAnswer:", | |
| "Question: Suppose that in a given FP Tree, an item in a leaf node N exists in every path. Which of the following is true?\n\nOptions:\nA. N co-occurs with its prefixes in every transaction\nB. {N}\u2019s minimum possible support is equal to the number of paths\nC. For every node P that is a parent of N in the FP tree, confidence(P->N) = 1\nD. The item N exists in every candidate set\n\nAnswer:", | |
| "Question: Tick the \\emph{false} assertion.\n\nOptions:\nA. The ElGamal encryption is non-deterministic.\nB. An ElGamal ciphertext is longer than an ElGamal plaintext.\nC. The security of ElGamal is based on the factorization problem.\nD. ElGamal uses public and private keys.\n\nAnswer:", | |
| "Question: Select \\emph{incorrect} statement. The brute force technique against a cipher with key $256$ bits is?\n\nOptions:\nA. impossible even if we can compute without burning an energy.\nB. impossible since the number of possible keys is too high $2^{256} \\approx 10^{77}$.\nC. impossible in future even if we consider Moore's law.\nD. feasible using all clusters at EPFL.\n\nAnswer:", | |
| "Question: Tick the \\emph{true} assertion related to the ElGamal signature scheme.\n\nOptions:\nA. A signature of a given message is obtained in a deterministic way.\nB. The generator $g$ generates a subgroup of prime order.\nC. The signature verification procedure is probabilistic.\nD. The main computational cost of the signature generation is due to one modular exponentiation.\n\nAnswer:", | |
| "Question: Which of AddressSanitizer (ASan), MemorySanitizer (MemSan),\n UndefinedBehaviorSanitizer (UBSan) or ThreadSanitizer (TSan) can detect\n bugs (if any) in the following code snippet?\n\nOptions:\nA. UBSan\nB. MemSan\nC. ASan\nD. TSan\n\nAnswer:", | |
| "Question: What is the gradient of $\\boldsymbol{x}^{\\top} \\boldsymbol{W}^{\\top} \\boldsymbol{W} \\boldsymbol{x}$ with respect to $\\boldsymbol{x}$ (written as a vector)?\n\nOptions:\nA. $2 \\boldsymbol{W}$\nB. $2 \\boldsymbol{W} \\boldsymbol{W}^{\\top} \\boldsymbol{x}$\nC. $2 \\boldsymbol{W}^{\\top} \\boldsymbol{x}$\nD. $2 \\boldsymbol{W}^{\\top} \\boldsymbol{W} \\boldsymbol{x}$\n\nAnswer:", | |
| "Question: Consider the Vernam Cipher. Let P=0010 1011 and K=1110 0110 and what is C=P $\\oplus$ K?\n\nOptions:\nA. 0011 0010\nB. 1100 0111\nC. 1100 1101\nD. 1101 0111\n\nAnswer:", | |
| "Question: In which of the following cases does JOS acquire the big kernel lock?\n\nOptions:\nA. Processor traps in user mode\nB. Processor traps in kernel mode\nC. Switching from kernel mode to user mode\nD. Initialization of application processor\n\nAnswer:", | |
| "Question: Tick the \\textbf{false} statement regarding the DES round function.\n\nOptions:\nA. There is an expansion operation $E$ from 32 to 48 bits.\nB. A round key is XORed to an internal register.\nC. There are $8$ identical S-boxes (substitution boxes) of size $6 \\times 4$.\nD. There is a permutation $P$ on 32-bits.\n\nAnswer:", | |
| "Question: Tick the \\emph{incorrect} assertion. The perfect cipher over $\\left\\{ 0,1 \\right\\}^\\ell$ is \\dots?\n\nOptions:\nA. uniformly distributed.\nB. $C^*\\;$.\nC. using a key space of size $2^\\ell!$.\nD. practical.\n\nAnswer:", | |
| "Question: Consider an array $A[1,\\ldots, n]$ consisting of the $n$ distinct numbers $1,2, \\ldots, n$. We are further guaranteed that $A$ is almost sorted in the following sense: $A[i] \\neq i$ for at most $\\sqrt{n}$ values of $i$. What are tight asymptotic worst-case running times for Insertion Sort and Merge Sort on such instances?\n\nOptions:\nA. It is $\\Theta(n \\log n)$ for Merge Sort and $\\Theta(n^2)$ for Insertion Sort.\nB. It is $\\Theta(n + \\sqrt{n}\\log n)$ for Merge Sort and $\\Theta(n)$ for Insertion Sort.\nC. It is $\\Theta(n + \\sqrt{n}\\log n)$ for Merge Sort and $\\Theta(n^2)$ for Insertion Sort.\nD. It is $\\Theta(n + \\sqrt{n}\\log n)$ for Merge Sort and $\\Theta(n^{3/2})$ for Insertion Sort.\n\nAnswer:", | |
| "Question: Which of the following statements are correct?\n\nOptions:\nA. Hinge loss (as in SVMs) is typically preferred over L2 loss (least squares loss) in classification tasks.\nB. Unions of convex sets are convex.\nC. One iteration of standard SGD for SVM costs roughly $\\Theta(D)$, where $D$ is the dimension.\nD. MSE (mean squared error) is typically more sensitive to outliers than MAE (mean absolute error).\n\nAnswer:", | |
| "Question: Consider:Non-terminals: S (top-level), NP (for \"noun phrase\"), VP (for \"verbal phrase\"), N (for \"Noun\"), V (for \"Verb\"), Det (for \"Determiner\").PoS tags: N, V, DetTerminals: I, yesterday, in, rain, went, home, the, cat, goOut of the following, select the ones which are possible valid \"syntactic rules\" as defined in a context-free grammar for processing (a tiny part of) English.A penalty will be applied for any incorrect answers.\n\nOptions:\nA. S \u2192 VP\nB. NP \u2192 Det N\nC. Det N \u2192 NP\nD. Det \u2192 went\n\nAnswer:", | |
| "Question: Which of AddressSanitizer (ASan), MemorySanitizer (MemSan),\n UndefinedBehaviorSanitizer (UBSan) or ThreadSanitizer (TSan) can detect\n bugs (if any) in the following code snippet?\n\nOptions:\nA. MemSan\nB. mentioned sanitizers can detect it.\nC. ASan\nD. There is at least one bug in the snippet, but none of the\n\nAnswer:", | |
| "Question: For logistic regression, what is the best encoding for the feature ``span'' which is the length of the bridge in meters? (One answer!!!!!!)?\n\nOptions:\nA. round to full meters\nB. find appropriate bins and use one-hot encoding\nC. find appropriate bins and use polynomial encoding\nD. logistic regression does not require an encoding\n\nAnswer:", | |
| "Question: Once software has been shipped, what does the Software Development Lifecycle require you to do to maintain security guarantees?\n\nOptions:\nA. Track the evolution of third party dependencies\nB. Deploy updates timely and safely\nC. Provide new features to attract new users\nD. Ensure the software works on newer machines\n\nAnswer:", | |
| "Question: Consider the following joint distribution on $X$ and $Y$, where both random variables take on the values $\\{0,1\\}: p(X=$ $0, Y=0)=0.1, p(X=0, Y=1)=0.2, p(X=1, Y=0)=0.3, p(X=1, Y=1)=0.4$. You receive $X=1$. What is the largest probability of being correct you can achieve when predicting $Y$ in this case?\n\nOptions:\nA. $\\frac{6}{7}$\nB. $\\frac{4}{7}$\nC. $\\frac{3}{7}$\nD. $\\frac{1}{4}$\n\nAnswer:", | |
| "Question: Which of the following is true regarding the random forest classification algorithm?\n\nOptions:\nA. It is not suitable for parallelization.\nB. It uses only a subset of features for learning in each weak learner.\nC. We compute a prediction by randomly selecting the decision of one weak learner.\nD. It produces a human interpretable model.\n\nAnswer:", | |
| "Question: Tick the \\emph{incorrect} statement. When $x\\rightarrow+\\infty$ \\ldots?\n\nOptions:\nA. $x^3 + 2x + 5 = \\mathcal{O}(x^3)$.\nB. $\\frac{1}{x^2} = \\mathcal{O}(\\frac{1}{x})$.\nC. $2^{\\frac{x}{\\log x}} = \\mathcal{O}(2^x)$.\nD. $n^x = \\mathcal{O}(x^n)$ for any constant $n>1$.\n\nAnswer:", | |
| "Question: Dan has been working with decision trees. His friend Eve recommends using random forests instead. What is most likely to happen?\n\nOptions:\nA. Accuracy will increase, interpretability will increase\nB. Accuracy will increase, interpretability will decrease\nC. Accuracy will decrease, interpretability will increase\nD. Accuracy will decrease, interpretability will decrease\n\nAnswer:", | |
| "Question: Which MAC construction is NOT based on the CBC mode?\n\nOptions:\nA. HMAC\nB. EMAC\nC. OMAC\nD. ISO/IEC 9797\n\nAnswer:", | |
| "Question: What is $(\\frac{9}{43})$?\n\nOptions:\nA. 1\nB. -1\nC. 0\nD. .209\n\nAnswer:", | |
| "Question: Categorical Cross-Entropy loss:?\n\nOptions:\nA. Minimizing the distance between the predicted point and the true point\nB. Maximizing the probability of the correct class\nC. Minimizing the score of false classes when they are close, or bigger than, the score of the true class\nD. Maximizing the accuracy\n\nAnswer:", | |
| "Question: Feature degeneracy:?\n\nOptions:\nA. when a set of variables in the feature space are not linearly independent\nB. when a set of variables in the feature space create chaotic results\nC. when a set of variables in the feature space have low impact on the output space\nD. when a point in output space can be obtained by one and only one set of variables in the feature space.\n\nAnswer:", | |
| "Question: What is true regarding Fagin's algorithm?\n\nOptions:\nA. It performs a complete scan over the posting files\nB. It provably returns the k documents with the largest aggregate scores\nC. Posting files need to be indexed by TF-IDF weights\nD. It never reads more than (kn)\u00bd entries from a posting list\n\nAnswer:", | |
| "Question: In Text Representation learning, which of the following statements is correct?\n\nOptions:\nA. Learning GloVe vectors can be done using SGD in a streaming fashion, by streaming through the input text only once.\nB. Every recommender systems algorithm for learning a matrix factorization $\\boldsymbol{W} \\boldsymbol{Z}^{\\top}$ approximating the observed entries in least square sense does also apply to learn GloVe word vectors.\nC. FastText performs unsupervised learning of word vectors.\nD. If you fix all word vectors, and only train the remaining parameters, then FastText in the two-class case reduces to being just a linear classifier.\n\nAnswer:", | |
| "Question: What is the content of the superblock in the JOS file system?\n\nOptions:\nA. List of all directories\nB. List of all files\nC. Magic number identifying the file system\nD. List of all blocks\n\nAnswer:", | |
| "Question: The Miller-Rabin test \\emph{cannot} output\\dots?\n\nOptions:\nA. ``prime'' for a ``composite'' number.\nB. ``prime'' for a ``prime'' number.\nC. ``composite'' for a ``composite'' number.\nD. ``composite'' for a ``prime'' number.\n\nAnswer:", | |
| "Question: In a first-preimage attack \\dots?\n\nOptions:\nA. you are given a hash function $h$ and you have to find two messages $m\\neq m'$ such that $h(m) = h(m')$.\nB. you are given a hash function $h$ and and a message $m$ and you have to have to find an $x$ such that $h(m)=x$.\nC. you are given a hash function $h$ and a message $m$ and you have to find $m'\\neq m$ such that $h(m) = h(m')$.\nD. you are given a hash function $h$ and a value $x$ and you have to have to find an $m$ such that $h(m)=x$.\n\nAnswer:", | |
| "Question: Which of the following is TRUE when comparing Vector Space Model (VSM) and Probabilistic Language Model (PLM)? (Slide 73 Week 2)?\n\nOptions:\nA. Both VSM and PLM require parameter tuning\nB. Both VSM and PLM use collection frequency in the model\nC. Both VSM and PLM take into account multiple term occurrences\nD. Both VSM and PLM are based on a generative language model\n\nAnswer:", | |
| "Question: The ElGamal cryptosystem is based on\\dots?\n\nOptions:\nA. nothing.\nB. the discrete logarithm problem.\nC. the RSA problem.\nD. the factorization problem.\n\nAnswer:", | |
| "Question: We define a simplified Gaussian Mixture Model consisting of 2 equally likely Gaussians, i.e. $K=2$ and $\\pi_1 = \\pi_2 = 0.5$, and covariance matrices of the form $\\mathbf{\\Sigma}_i = \\sigma_i \\mathbf{I}_{D\times D}$ for $i\\in \\{1, 2\\}$ with $\\mathbf{I}_{D\times D}$ the identity matrix of size $D$. The dataset consists of only 2 points $\\xv_1$ and $\\xv_2$ that are distinct ($\\xv_1 \not= \\xv_2$). We initialize the model at some finite $\boldsymbol{\\mu}_1^{(0)}, \boldsymbol{\\mu}_2^{(0)}$ and $\\sigma_1^{(0)}, \\sigma_2^{(0)}$. We fit the model by the EM method on these parameters (keeping $\\pi_1$ and $\\pi_2$ fixed to 0.5). After $T\to \\infty$ steps, select the true statement among the following: ?\n\nOptions:\nA. $\\sigma_1^{(T)}$ and $\\sigma_2^{(T)}$ converge to 0 for any initializations. \nB. $\\sigma_1^{(T)}$ and $\\sigma_2^{(T)}$ diverge to $\\infty$ for any initializations.\nC. $\\sigma_1^{(T)}$ and $\\sigma_2^{(T)}$ converge to 0 for some but not all the initializations.\nD. $\\sigma_1^{(T)}$ and $\\sigma_2^{(T)}$ diverge to $\\infty$ for some but not all the initializations.\n\nAnswer:", | |
| "Question: Which of the following statements about mitigations are true?\n\nOptions:\nA. The bug remains in the application, mitigations simply make exploitation harder\nB. The performance of certain mitigations depends on underlying architecture features (e.g., i386 versus x86-64)\nC. All mitigations fully stop an attack vector\nD. No mitigation requires hardware support to be implemented\n\nAnswer:", | |
| "Question: Given a message $x$, find a different message $x'$ such that $h(x)=h(x')$ is the definition of ...\n\nOptions:\nA. First preimage attack.\nB. Collision attack.\nC. Second preimage attack.\nD. Known message attack.\n\nAnswer:", | |
| "Question: Graph coloring is \\ldots?\n\nOptions:\nA. \\ldots $NP$-hard with 1 color.\nB. \\ldots not interesting for cryptographers.\nC. \\ldots an $NP$-complete problem when limited to 3 colors.\nD. \\ldots always possible with 2 colors.\n\nAnswer:", | |
| "Question: Let $n$ be an integer such that $n\\geq 2$ and let $A \\in \\R^{n\times n}$, and $xv \\in \\R^n$, consider the function $f(xv) = xv^\top A xv$ defined over $\\R^n$. Which of the following is the gradient of the function $f$? ?\n\nOptions:\nA. $2 xv^\top A$\nB. $2Axv$\nC. $A^\top xv + Axv$\nD. $2A^\top xv$\n\nAnswer:", | |
| "Question: Mean Square Error loss:?\n\nOptions:\nA. Minimizing the distance between the predicted point and the true point\nB. Maximizing the probability of the correct class\nC. Minimizing the score of false classes when they are close, or bigger than, the score of the true class\nD. Maximizing the accuracy\n\nAnswer:", | |
| "Question: How do you split your data? (One or multiple answers)?\n\nOptions:\nA. 60\\% of the oldest bridges into training, the next 20\\% into validation and the newest 20\\% of bridges into the test set\nB. randomly sample with a $60\\%:20\\%:20\\%$ split for training, validation and test set\nC. use steel bridges for training and concrete bridges for the test set\nD. use long bridges for training and short bridges for the test set\n\nAnswer:", | |
| "Question: Suppose that q is density reachable from p. The chain of points that ensure this relationship are {t,u,g,r}. Which of the following is always true?\n\nOptions:\nA. p is density reachable from q\nB. q and p are density-connected\nC. p is a border point\nD. q is a core point\n\nAnswer:", | |
| "Question: If A={a} and B={b}, select all strings that belongs to\u00a0(A \u2297 B)+\nA penalty will be applied for any wrong answers selected.\n\nOptions:\nA. (a,b)\nB. (aaa,bbb)\nC. (a,bb)\nD. (aaaaa,bbbb)\n\nAnswer:", | |
| "Question: A neural network has been trained for multi-class classification using cross-entropy but has not necessarily achieved a global or local minimum on the training set.\n\tThe output of the neural network is $\\mathbf{z}=[z_1,\\ldots,z_d]^\top$ obtained from the penultimate values $\\mathbf{x}=[x_1,\\ldots,x_d]^\top$ via softmax $z_k=\frac{\\exp(x_k)}{\\sum_{i}\\exp(x_i)}$ that can be interpreted as a probability distribution over the $d$ possible classes.\n\tThe cross-entropy is given by $H(\\mathbf{y},\\mathbf{z})=-\\sum_{i=1}^{d} y_i \\ln{z_i}$ where $\\mathbf{y}$ is one-hot encoded meaning the entity corresponding to the true class is 1 and other entities are 0.\n\nOptions:\nA. One transformation has no effect, the other sometimes increases and sometimes decreases the accuracy.\nB. Neither transformation affects the accuracy.\nC. One transformation has no effect, the other one decreases the accuracy in some cases (but never increases it).\nD. Both transformations decrease the accuracy in some cases (but never increase it).\n\nAnswer:", | |
| "Question: The Shannon theorem states that perfect secrecy implies...\n\nOptions:\nA. $H(K)=H(X)$\nB. $H(Y)\\geq H(X)$\nC. $H(K)\\geq H(X)$\nD. $H(Y)\\leq H(X)$\n\nAnswer:", | |
| "Question: Tick the \\textbf{true} assertion. Let $X$ be a random variable that is equal to zero with probability 1/2 and to 1 with probability 1/2. Since the variance of $X$ is $V(X)=E((X-E(X))^2)= E(X^2)-(E(X))^2$, here $V(X)$ is:?\n\nOptions:\nA. 1/4\nB. 1/2\nC. 1/3\nD. 1/8\n\nAnswer:", | |
| "Question: Data being classified as unstructured or structured depends on the:?\n\nOptions:\nA. Degree of abstraction\nB. Level of human involvement\nC. Type of physical storage\nD. Amount of data\n\nAnswer:", | |
| "Question: Let P be the statement \u2200x(x>-3 -> x>3). Determine for which domain P evaluates to true: ?\n\nOptions:\nA. -3<x<3\nB. x>-3\nC. x>3\nD. None of the other options\n\nAnswer:", | |
| "Question: Which of the following statement(s) is/are true about Safe Exception Handling (SEH)?\n\nOptions:\nA. Neither SafeSEH nor SeHOP checks the order and number of exception handlers.\nB. The implementation of SEH is compiler specific.\nC. SEH is a defense that protects C/C++ programs against control-flow hijack attacks through changing exception data structures.\nD. SafeSEH provides stronger protection than SeHOP.\n\nAnswer:", | |
| "Question: Which of the following Diffie-Hellman instance is believed to be secure?\n\nOptions:\nA. Diffie-Hellman in a subgroup of $\\left\\{ 0,\\dots,n \\right\\}$ (with the addition) of prime order $q$ with $q$ a $200$-bit prime and $n$ a $2048$-bit integer.\nB. Diffie-Hellman over a subgroup of a good Elliptic curve over $Z_p$ of prime order $q$, with $q$ a $200$-bit prime and $p$ a $2048$-bit prime.\nC. Diffie-Hellman over a subgroup of $Z_p^*$ of order $q$, with $q$ a $30$-bit prime and $p$ a $2048$-bit prime.\nD. Diffie-Hellman over a subgroup of $Z_p^*$ of order $q=2^{128}(127)^{40}$, with $p$ a $2048$-bit prime.\n\nAnswer:", | |
| "Question: The output feedback (OFB) mode of operation \\dots?\n\nOptions:\nA. requires its nonce to be public and constant to be secure.\nB. requires its nonce IV to be constant but secret to be secure.\nC. requires its nonce to be new for every plaintext to be secure.\nD. doesn't require any nonce.\n\nAnswer:", | |
| "Question: Recall that the hard-margin SVM problem corresponds to:\n\t$$ \\underset{\\substack{\\ww \\in \\R^d, \\ \forall i:\\ y_i \\ww^\top \\xx_i \\geq 1}}{\\min} \\Vert \\ww \\Vert_2.$$\n\tNow consider the $2$-dimensional classification dataset corresponding to the $3$ following datapoints: $\\xx_1 = (-1, 2)$, $\\xx_2 = (1, 2)$, $\\xx_3 = (0, -2)$ and $y_1 = y_2 = 1$, $y_3 = -1$.\n\tWhich of the following statements is \textbf{true}:\n ?\n\nOptions:\nA. Our dataset is not linearly separable and hence it does not make sense to consider the hard-margin problem.\nB. There exists a unique $\\ww^\\star$ which linearly separates our dataset.\nC. The unique vector which solves the hard-margin problem for our dataset is $\\ww^\\star = (0, 1)$.\nD. None of the other statements are true.\n\nAnswer:", | |
| "Question: In which of the following cases does the TLB need to be flushed?\n\nOptions:\nA. Inserting a new page into the page table for a user-space application.\nB. Deleting a page from the page table.\nC. Changing the read/write permission bit in the page table.\nD. Inserting a new page into the page table for kernel.\n\nAnswer:", | |
| "Question: Which of the following is true?\n\nOptions:\nA. High precision implies low recall\nB. High precision hurts recall\nC. High recall hurts precision\nD. High recall implies low precision\n\nAnswer:", | |
| "Question: Which adversarial model corresponds to Linear Cryptanalysis?\n\nOptions:\nA. Chosen Ciphertext Attack\nB. Chosen Plaintext Attack\nC. Known Ciphertext Attack\nD. Known Plaintext Attack\n\nAnswer:", | |
| "Question: Which of the following elements belong to $\\mathbf{Z}_{35}^*$?\n\nOptions:\nA. $12$.\nB. $20$.\nC. $14$.\nD. $15$.\n\nAnswer:", | |
| "Question: The Time-Memory Tradeoff Attack ...\n\nOptions:\nA. is useful for finding a preimage within complexity $O\\big(\\big({\\frac{2}{3}}\\big)^N\\big).$\nB. is useful for finding a preimage within complexity $O(N^{\\frac{2}{3}}).$\nC. is a dedicated method which works only on SHA1.\nD. can be combined with birthday paradox to find the order of the group in RSA efficiently.\n\nAnswer:", | |
| "Question: Maintaining the order of document identifiers for vocabulary construction when partitioning the document collection is important?\n\nOptions:\nA. in the index merging approach for single node machines\nB. in the map-reduce approach for parallel clusters\nC. in both\nD. in neither of the two\n\nAnswer:", | |
| "Question: After the join step, the number of k+1-itemsets ?\n\nOptions:\nA. is equal to the number of frequent k-itemsets\nB. can be equal, lower or higher than the number of frequent k-itemsets\nC. is always higher than the number of frequent k-itemsets\nD. is always lower than the number of frequent k-itemsets\n\nAnswer:", | |
| "Question: Consider a binary classification problem with classifier $f(\\mathbf{x})$ given by $$ f(\\mathbf{x})= \\begin{cases}1, & g(\\mathbf{x}) \\geq 0 \\\\ -1, & g(\\mathbf{x})<0\\end{cases} $$ and $\\mathbf{x} \\in \\mathbb{R}^{6}$. Consider a specific pair $(\\mathbf{x}, y=1)$ and assume that $g(\\mathbf{x})=8$. In particular this means that this point is classified correctly by $f$. Assume further that we have computed the gradient of $g$ at $\\mathbf{x}$ to be $\\nabla_{\\mathbf{x}} g(\\mathbf{x})=(+1,-2,+3,-4,+5,-6)$. You are allowed to make one step in order to (hopefully) find an adversarial example. In the following four questions, assume $\\epsilon=1$. Which offset $\\delta$ with $\\|\\delta\\|_{\\infty} \\leq 1$ yields the smallest value for $g(\\mathbf{x}+\\delta)$, assuming that $g$ is (locally) linear?\n\nOptions:\nA. $(-1,-1,-1,-1,-1,-1)$\nB. $(+1,+1,+1,+1,+1,+1)$\nC. $(-1,+2,-3,+4,-5,+6)$\nD. $(-1,+1,-1,+1,-1,+1)$\n\nAnswer:", | |
| "Question: Consider the following shellcode, which of the following statement(s) is/are correct?\n\n\\begin{lstlisting}[language=nasm,style=nasm]\nneedle: jmp gofar\ngoback: pop %rdi\n \txor %rax, %rax\n \tmovb $0x3b, %al\n \txor %rsi, %rsi\n \txor %rdx, %rdx\n \tsyscall\ngofar: call goback\n.string \"/bin/sh\"\n\\end{lstlisting}?\n\nOptions:\nA. Lines 2-6 are preparing arguments for the syscall invocation.\nB. In the exploit payload, the string \"/bin/sh\" must end with a \"0x0\" byte to ensure it is terminated correctly.\nC. The purpose of line 8 is to push the address of \"/bin/sh\" to the stack and jump to line 2.\nD. Line 3 is not necessary.\n\nAnswer:", | |
| "Question: Tick the \\emph{correct} assertion.\n\nOptions:\nA. The only way of finding a collision in a hash function while using (almost) no memory is exhaustive search.\nB. Floyd's cycle finding algorithm returns the closest bike shop.\nC. Exhaustive search is always a concern in cryptographic schemes.\nD. $\\mathbb{Z}_p^*$ is a field.\n\nAnswer:", | |
| "Question: With negative sampling a set of negative samples is created for\n?\n\nOptions:\nA. For each occurrence of a word in the text\nB. For each word of the vocabulary\nC. For each word-context pair\nD. For each occurrence of a word-context pair in the text\n\nAnswer:", | |
| "Question: Daemons are just long running processes. When applying mitigations to these\n processes, several aspects change. Which ones?\n\nOptions:\nA. ASLR becomes less effective as multiple requests across\nB. handled by the same thread.\nC. different users are handled in a single process.\nD. Stack canaries become less effective as multiple requests are\n\nAnswer:", | |
| "Question: Select the morpho-syntactic categories that do not carry much semantic content and are thus usually filtered-out from indexing.\n\nOptions:\nA. Adjectives\nB. Conjunctions\nC. Determiners\u00a0\nD. Nouns\n\nAnswer:", | |
| "Question: The differential probability of a function $f:\\{0,1\\}^p\\rightarrow \\{0,1\\}^q$ is, given $a\\in\\{0,1\\}^p$ and $b\\in\\{0,1\\}^q$, defined by \\dots?\n\nOptions:\nA. $\\mathrm{DP}^f(a,b)=\\Pr_{X\\in _U \\{0,1\\}^p} [f(X\\oplus a)=f(X\\oplus b)] $.\nB. $\\mathsf{DP}^f(a,b)=\\Pr_{X\\in _U \\{0,1\\}^p} [f(X)\\oplus a=f(X)\\oplus b] $.\nC. $\\mathsf{DP}^f(a,b)=\\Pr_{X\\in _U \\{0,1\\}^p} [f(X\\oplus b)=f(X)\\oplus a] $.\nD. $\\mathsf{DP}^f(a,b)=\\Pr_{X\\in _U \\{0,1\\}^p} [f(X\\oplus a)=f(X)\\oplus b] $.\n\nAnswer:", | |
| "Question: How many different simple substitution ciphers do exist with respect to an alphabet of 26 characters?\n\nOptions:\nA. $26!$\nB. $2^{26}$\nC. $26^2$\nD. 26\n\nAnswer:", | |
| "Question: Which cryptographic primitive(s) is (are) used in S/Key - OTP ?\n\nOptions:\nA. Only encryption and a hash function\nB. Only encryption and a MAC algorithm\nC. Only a hash function\nD. Only a MAC\n\nAnswer:", | |
| "Question: Commitments should \\ldots?\n\nOptions:\nA. resist forgery.\nB. be binding and hiding.\nC. allow a third party to open commitments.\nD. resist preimage attacks.\n\nAnswer:", | |
| "Question: Which of the following is TRUE for Recommender Systems (RS)?\n\nOptions:\nA. The complexity of the Content-based RS depends on the number of users\nB. Item-based RS need not only the ratings but also the item features\nC. Matrix Factorization is typically robust to the cold-start problem.\nD. Matrix Factorization can predict a score for any user-item combination in the dataset.\n\nAnswer:", | |
| "Question: Which of the following is/are typical AddressSanitizer use cases?\n\nOptions:\nA. Out-of-bounds violations visibility increase during fuzzing\nB. Use-after-free mitigation in production\nC. Uninitialized variable detection during unit tests\nD. NULL-pointer dereference detection during fuzzing\n\nAnswer:", | |
| "Question: A constant initialization vector (IV), i.e., same for every plaintext, can be used in the ...\n\nOptions:\nA. CTR mode.\nB. CBC mode.\nC. OFB mode.\nD. ECB mode.\n\nAnswer:", | |
| "Question: In a bruteforce attack on $N$ keys, when the distribution $D$ of the keys is \\emph{known} and when \\emph{all keys have a non-zero probability to occur},\\dots?\n\nOptions:\nA. the best strategy is to try each key uniformly at random.\nB. the best strategy is to sort each key by decreasing order of likelihood and try them.\nC. the \\emph{worst} case complexity depends on the distribution $D$.\nD. the \\emph{expected} complexity when using the \\emph{best strategy} is $(N+1)/2$ for all distributions $D$.\n\nAnswer:", | |
| "Question: What is the \\textbf{minimal} requirement to make the Vigen\\`ere cipher perfectly secure.\n\nOptions:\nA. Vigen\\`ere is always perfectly secure.\nB. Changing the key after each encryption is enough.\nC. Having a key that has the same length as the plaintext is enough.\nD. We need to use a uniformly distributed fresh key at each encryption that has the same length as the plaintext.\n\nAnswer:", | |
| "Question: KEM \\dots?\n\nOptions:\nA. stands for Keyless Encryption Mechanism.\nB. is a Korean encryption mechanism.\nC. is a symmetric-key algorithm.\nD. is a public-key algorithm.\n\nAnswer:", | |
| "Question: Tick the \\textit{incorrect} assertion.\n\nOptions:\nA. Using elliptic curves, it is possible to obtain a secure 3-party Diffie-Hellman key-exchange.\nB. Cryptographic primitives based on elliptic curves are popular because in general, they require smaller primes than other public-key cryptosystems to be secure.\nC. A non-singular elliptic curve $E_{a,b}(\\mathrm{GF}(p^k))$, with $p>3$ a prime and $k$ a positive integer forms a field.\nD. Standardized random curves over prime fields are believed to be secure.\n\nAnswer:", | |
| "Question: Which of the following models for generating vector representations for text require to precompute the frequency of co-occurrence of words from the vocabulary in the document collection?\n\nOptions:\nA. LSI\nB. CBOW\nC. Fasttext\nD. Glove\n\nAnswer:", | |
| "Question: Select \\emph{incorrect} statement. Brithday paradox?\n\nOptions:\nA. is a brute force technique.\nB. can be implemented with constant memory using Rho ($\\rho$) method.\nC. is used to recover the secret key of AES in $2^{64}$ computations.\nD. can be implemented using a table of size $\\Theta\\sqrt{N}$\n\nAnswer:", | |
| "Question: Tick the \\textbf{false} assertion. In an interactive proof,?\n\nOptions:\nA. \\ldots the completeness property does depend on the random coins of the prover.\nB. \\ldots the complexity of a dishonest prover is kept within polynomial bounds.\nC. \\ldots the soundness property places a threshold on the acceptance of illegitimate proofs.\nD. \\ldots the complexity of the verifier is polynomial in the size of the input.\n\nAnswer:", | |
| "Question: For an interactive proof system, the difference between perfect, statistical and computational zero-knowledge is based on \\ldots?\n\nOptions:\nA. \\ldots the distinguishability between some distributions.\nB. \\ldots the percentage of recoverable information from a transcript with a honest verifier.\nC. \\ldots the number of times the protocol is run between the prover and the verifier.\nD. \\ldots whether the inputs are taken in $\\mathcal{P}$, $\\mathcal{NP}$ or $\\mathcal{IP}$.\n\nAnswer:", | |
| "Question: Finding collisions on a set of N elements ...\n\nOptions:\nA. requires the storage of size $ \\Omega(N).$\nB. requires time $ O({N}^{\\frac{1}{3}}).$\nC. can be done with the storage of size $O(1).$\nD. is doable for $N=2^{256}$.\n\nAnswer:", | |
| "Question: Let $C$ be a perfect cipher with $\\ell$-bit blocks. Then, \\dots?\n\nOptions:\nA. for $x_1 \\neq x_2$, $\\Pr[C(x_1) = y_1, C(x_2)=y_2] = \\frac{1}{2^{2\\ell}}$.\nB. the size of the key space of $C$ should be at least $(2^{\\ell}!)$.\nC. given pairwise independent inputs to $C$, the corresponding outputs are independent and uniformly distributed.\nD. $C$ has an order $3$ decorrelation matrix which is equal to the order $3$ decorrelation matrix of a random function.\n\nAnswer:", | |
| "Question: Let $f: \\mathbb{Z}_{m n} \\rightarrow \\mathbb{Z}_m \\times \\mathbb{Z}_n$ be defined by $f (x) = (x \\bmod m,x \\bmod n)$. Then $f$ is a ring isomorphism between $\\mathbb{Z}_{180}$ and:?\n\nOptions:\nA. $\\mathbb{Z}_{2} \\times \\mathbb{Z}_{90}$.\nB. $\\mathbb{Z}_{4} \\times \\mathbb{Z}_{45}$.\nC. $\\mathbb{Z}_{10} \\times \\mathbb{Z}_{18}$.\nD. $\\mathbb{Z}_{6} \\times \\mathbb{Z}_{30}$.\n\nAnswer:", | |
| "Question: The Murphy Law states that if there is a single security hole in an exposed cryptosystem, then\\dots?\n\nOptions:\nA. hope for the best\nB. nobody will look for it\nC. nobody will find it\nD. someone will ultimately find it\n\nAnswer:", | |
| "Question: Which one of these is \\emph{not} a MAC construction?\n\nOptions:\nA. HMAC\nB. OMAC\nC. MD5\nD. ISO/IEC 9797\n\nAnswer:", | |
| "Question: Let S(x) be the statement \u201cx has been in a lake\u201d and L(x) be the statement \u201cx lives in Lausanne\u201d and the domain of x consists of all the humans in the world.\n\nOptions:\nA. L(x)) \\wedge \forall y \\left[ \nB. \\( \\exists x \\Bigr[( S(x) \\wedge \nC. \\( \\exists! x (S(x) \\wedge L(x)) \\)\nD. S(y) \\wedge \n\nAnswer:", | |
| "Question: In the first pass over the database of the FP Growth algorithm?\n\nOptions:\nA. Frequent itemsets are extracted\nB. A tree structure is constructed\nC. The frequency of items is computed\nD. Prefixes among itemsets are determined\n\nAnswer:", | |
| "Question: Consider a MAC defined by $\\mathsf{MAC}: \\{0,1\\}^* \\times \\{0,1\\}^k \\mapsto \\{0,1\\}^n$. The complexity of a generic key recovery attacks against $\\mathsf{MAC}$ is \\ldots?\n\nOptions:\nA. $2^{k/2}$\nB. $2^k$\nC. $2^{n/2}$\nD. $2^n$\n\nAnswer:", | |
| "Question: Which assumption(s) is (are) required during the exchange of a key which will be used by a block cipher?\n\nOptions:\nA. Nothing.\nB. Authentication.\nC. Authentication and integrity.\nD. Confidentiality.\n\nAnswer:", | |
| "Question: The following function(s) have a unique minimizer.\n\nOptions:\nA. (a) $f(x)=x^{2}, x \\in[-3,2]$\nB. (b) $f(x)=\\log (x), x \\in(0,10]$\nC. (c) $f(x)=\\sin (x), x \\in[-10,10]$\nD. (d) $f(x)=e^{3 x}+x^{4}-3 x, x \\in[-10,10]$\n\nAnswer:", | |
| "Question: Which of the following statements about coverage-guided fuzzing is/are correct?\n\nOptions:\nA. Counting the number of times the covered code has been executed\nB. provides a more fine-grained view of program behavior than only\nC. Redundant seeds in the corpus will reduce fuzzing efficiency.\nD. \"covered/not covered\" binary code coverage.\n\nAnswer:", | |
| "Question: Let $H:\\{0,1\\}^* \\rightarrow \\{0,1\\}^n$ be a hash function and $x_1,x_2\\in\\{0,1\\}^{2n}$ two random different messages. In cryptography, we usually assume that the probability of collision, i.e. $\\Pr[H(x_1)=H(x_2)]$, is close to \\ldots?\n\nOptions:\nA. $2^{-\\frac{n}{2}}$.\nB. $2^{-n}$.\nC. $2^{-2n}$.\nD. $0$.\n\nAnswer:", | |
| "Question: Consider the following joint distribution on $X$ and $Y$, where $X \\in\\{-1,0,1\\}$ and $Y \\in\\{0,1\\}: p(X=-1, Y=0)=0.05$, $p(X=-1, Y=1)=0.05, p(X=0, Y=0)=0.1, p(X=0, Y=1)=0.1, p(X=1, Y=0)=0.3, p(X=1, Y=1)=$ 0.4. You learn that $X \\geq 0$. What is the largest probability of being correct you can achieve when predicting $Y$ in this case?\n\nOptions:\nA. $\\frac{4}{7}$\nB. $\\frac{5}{9}$\nC. $0$\nD. $\\frac{6}{7}$\n\nAnswer:", | |
| "Question: What is the algorithm to perform optimization with gradient descent? Actions between Start loop and End loop are performed multiple times. (One answer)?\n\nOptions:\nA. 1 Start loop, 2 Initialize weights, 3 Compute gradients, 4 Update weights, 5 End loop\nB. 1 Initialize weights, 2 Compute gradients, 3 Start loop, 4 Update weights, 5 End loop\nC. 1 Initialize weights, 2 Start loop, 3 Update weights, 4 End loop, 5 Compute gradients \nD. 1 Initialize weights, 2 Start loop, 3 Compute gradients, 4 Update weights, 5 End Loop\n\nAnswer:", | |
| "Question: Which of the following is/are true about fuzzing?\n\nOptions:\nA. no longer improving coverage\nB. The efficacy of a fuzzing campaign scales with its speed (executions per second)\nC. Fuzzers generally determine the exploitability of a crash.\nD. Fuzzers may get ``stuck'' and cannot easily detect that they are\n\nAnswer:", | |
| "Question: What is the mean squared error of $f$ for a sample, where $\\textbf{x}$ is an input, $y$ a target and $f(\\textbf{x},W)$ the mapping function ? \n(One answer)?\n\nOptions:\nA. $||y - f(\\textbf{x},W)||^2 $ \nB. $||y - f(\\textbf{x},W)|| $\nC. $-\\log(P(y=i | \\textbf{x})) = -\\log(\\frac{e^{\\textbf{f}_i(\\textbf{x},W)}}{\\sum_j e^{\\textbf{f}_j(\\textbf{x},W)}})$ \nD. $P(y=i |\\textbf{x}) = \\frac{e^{\\textbf{f}_i(\\textbf{x},W)}}{\\sum_j e^{\\textbf{f}_j(\\textbf{x},W)}}$ \n\nAnswer:", | |
| "Question: Compared to the plain RSA cryptosystem and for equivalent key sizes, the plain Elgamal cryptosystem has\\dots?\n\nOptions:\nA. a simpler key generation algorithm.\nB. a simpler encryption algorithm.\nC. a simpler decryption algorithm.\nD. shorter ciphertexts.\n\nAnswer:", | |
| "Question: Which of the following apply to recent Android-based mobile systems but not\n to Linux-based desktop systems?\n\nOptions:\nA. Apps should use the binder interface to communicate with other\nB. directories.\nC. Arbitrary apps can exchange files through shared\nD. calls available.\n\nAnswer:", | |
| "Question: Tick the \\textbf{false} assertion. The Advantage can be computed \\ldots?\n\nOptions:\nA. \\ldots with a differential characteristic.\nB. \\ldots as a distance.\nC. \\ldots with a frequentist approach.\nD. \\ldots with a Bayesian approach.\n\nAnswer:", | |
| "Question: The Kerckhoffs Principle states that ...\n\nOptions:\nA. the security of a cryptosystem should be based on an NP-hard problem.\nB. all ciphertexts appear with the same probability.\nC. the adversary may know the description of a cryptosystem.\nD. the cryptosystem should be public.\n\nAnswer:", | |
| "Question: Consider the following mysterious binary encoding:\begin{center} \begin{tabular}{c|c} symbol & encoding \\ \\hline $a$ & $??0$\\ $b$ & $??0$\\ $c$ & $??0$\\ $d$ & $??0$ \\end{tabular} \\end{center} where with '$?$' we mean that we do not know which bit is assigned as the first two symbols of the encoding of any of the source symbols $a,b,c,d$. What can you infer on this encoding assuming that the code-words are all different?\n\nOptions:\nA. The encoding is uniquely-decodable.\nB. The encoding is uniquely-decodable but not prefix-free.\nC. We do not possess enough information to say something about the code.\nD. It does not satisfy Kraft's Inequality.\n\nAnswer:", | |
| "Question: Select all valid answers about UNIX-like shell.\n\nOptions:\nA. The shell is a program, that runs in kernel-space.\nB. The shell is a user interface for UNIX-like systems.\nC. The shell is a program, which reads from standard input.\nD. The shell is a program, that runs in user-space.\n\nAnswer:", | |
| "Question: Modularity of a social network always:?\n\nOptions:\nA. Increases with the number of communities\nB. Increases when an edge is added between two members of the same community\nC. Decreases when new nodes are added to the social network that form their own communities\nD. Decreases if an edge is removed\n\nAnswer:", | |
| "Question: 10 itemsets out of 100 contain item A, of which 5 also contain B. The rule B -> A has:?\n\nOptions:\nA. unknown support and 50% confidence\nB. unknown support and unknown confidence\nC. 5% support and 50% confidence\nD. 5% support and unknown confidence\n\nAnswer:", | |
| "Question: Consider the following joint distribution on $X$ and $Y$, where both random variables take on the values $\\{0,1\\}: p(X=$ $0, Y=0)=0.1, p(X=0, Y=1)=0.2, p(X=1, Y=0)=0.3, p(X=1, Y=1)=0.4$. You receive $X=1$. What is the largest probability of being correct you can achieve when predicting $Y$ in this case?\n\nOptions:\nA. $\\frac{4}{7}$\nB. $1$\nC. $\\frac{2}{3}$\nD. $0$\n\nAnswer:", | |
| "Question: Given graph 1\u21922, 1\u21923, 2\u21923, 3\u21922, switching from Page Rank to Teleporting\u00a0PageRank will have an influence on the value(s) of:?\n\nOptions:\nA. All the nodes\nB. Node 1\nC. Node 2 and 3\nD. No nodes. The values will stay unchanged.\n\nAnswer:", | |
| "Question: Tick the \\textbf{true} assertion. A Universal Forgery is \\ldots?\n\nOptions:\nA. \\ldots a forgery where the adversary is able to forge a valid MAC/signature for an arbitrary message.\nB. \\ldots a forgery where the adversary is able to forge a valid MAC/signature for a new message.\nC. \\ldots a forgery where the adversary has to recover the secret key.\nD. \\ldots a forgery where the adversary plays chess.\n\nAnswer:", | |
| "Question: Consider:Non-terminals: S (top-level), NP (for \"noun phrase\"), VP (for \"verbal phrase\"), N (for \"Noun\"), V (for \"Verb\"), Det (for \"Determiner\").PoS tags: N, V, DetTerminals: I, yesterday, in, rain, went, home, the, cat, goOut of the following, select the ones which are possible valid \"syntactic rules\" as defined in a context-free grammar for processing (a tiny part of) English.A penalty will be applied for any incorrect answers.\n\nOptions:\nA. NP \u2192 Det N\nB. VP \u2192 the cat\nC. VP \u2192 VP NP\nD. VP NP \u2192 V N\n\nAnswer:", | |
| "Question: In RSA, we use CRT ...\n\nOptions:\nA. to speedup encryption.\nB. to speedup decryption.\nC. since it is necessary operation of the primitive.\nD. to prove security.\n\nAnswer:", | |
| "Question: Which of the following sets can you use Cantor\u2019s Diagonalization Argument to prove it is uncountable (multiple answers) ?\n\nOptions:\nA. Real numbers not containing 7 in their decimal representation.\nB. Real numbers between 1 and 2.\nC. The set binary strings with finite length (i.e. 01, 10, 010, ...)\nD. The set of decimal representations of negative real numbers .\n\nAnswer:", | |
| "Question: Which of the following statements about testing is/are correct?\n\nOptions:\nA. Concolic execution specializes in finding bugs close to the path\nB. indirect calls is typically challenging but required to minimize\nC. In static analysis, determining the correct target set of\nD. of the provided concrete input.\n\nAnswer:", | |
| "Question: In terms of the \\textbf{bias-variance} decomposition, a 1-nearest neighbor classifier has \\rule{2cm}{0.15mm} than a 3-nearest neighbor classifier.\n\nOptions:\nA. higher variance\nB. lower variance\nC. higher bias\nD. lower bias\n\nAnswer:", | |
| "Question: When representing the adjacency list of a Web page in a connectivity server by using a reference list from another Web page, the reference list is searched only in a neighboring window of the Web page's URL, because:?\n\nOptions:\nA. Subsequent URLs in an adjacency list have typically small differences\nB. Typically many URLs in a web page are similar to each other\nC. Often many URLs among two pages with similar URL are similar\nD. Most extra nodes are found in the neighboring window\n\nAnswer:", | |
| "Question: Which of the following measures will always improve fuzzing executions per second?\n\nOptions:\nA. Reducing overheads imposed by the fuzzing framework.\nB. Performing structure-aware input generation.\nC. Providing dictionaries for input generation.\nD. Collecting code coverage as feedback.\n\nAnswer:", | |
| "Question: What is the complexity of generating an RSA modulus of length $2\\ell$?\n\nOptions:\nA. $O(\\ell)$\nB. $O(\\ell^2)$\nC. $O(\\ell^4)$\nD. $O(\\ell^8)$\n\nAnswer:", | |
| "Question: Which of the following statements about the $\\mathrm{SVD}$ of an $N \\times D$ matrix $\\mathbf{X}$ are correct?\n\nOptions:\nA. We can compute the singular values of $\\mathbf{X}$ by computing the eigenvalues of $\\mathbf{X X}^{\\top}$. This has complexity $O\\left(D^{3}\\right)$.\nB. We can compute the singular values of $\\mathbf{X}$ by computing the eigenvalues of $\\mathbf{X X}^{\\top}$. This has complexity $O\\left(N^{3}\\right)$.\nC. We can compute the singular values of $\\mathbf{X}$ by computing the eigenvalues of $\\mathbf{X}^{\\top} \\mathbf{X}$. This has complexity $O\\left(D^{3}\\right)$.\nD. We can compute the singular values of $\\mathbf{X}$ by computing the eigenvalues of $\\mathbf{X}^{\\top} \\mathbf{X}$. This has complexity $O\\left(N^{3}\\right)$.\n\nAnswer:", | |
| "Question: Bluetooth pairing v2.0 is based on\\dots?\n\nOptions:\nA. bilinar mappings over elliptic curves.\nB. a short authenticated string.\nC. an ephemeral secret PIN code.\nD. a secure token.\n\nAnswer:", | |
| "Question: The Factoring Problem is equivalent to ...\n\nOptions:\nA. the Computing Square Roots Problem.\nB. the RSA Decryption Problem.\nC. the Discrete Logarithm Problem.\nD. the Computing Generator Problem.\n\nAnswer:", | |
| "Question: Full-text retrieval refers to the fact that?\n\nOptions:\nA. grammatical variations of a word are considered as the same index terms\nB. the document text is grammatically fully analyzed for indexing\nC. queries can be formulated as texts\nD. all words of a text are considered as potential index terms\n\nAnswer:", | |
| "Question: When using the plain ElGamal cryptosystem over the group $\\mathbb{Z}_p^*$ with a fixed $\\ell$-bit prime $p$ and a fixed generator $g\\in\\mathbb{Z}_p^*$, the \\textit{tightest} complexity of generating a new key-pair is\\ldots?\n\nOptions:\nA. $\\mathcal{O}(4{\\ell})$.\nB. $\\mathcal{O}(\\ell^3)$.\nC. $\\mathcal{O}(\\ell^2)$.\nD. $\\mathcal{O}(\\log{\\ell}^4)$.\n\nAnswer:", | |
| "Question: One-time pad ...\n\nOptions:\nA. never uses a key $K$ which is picked from a uniform distribution.\nB. pads the message at least once before encryption.\nC. allows an efficient key management.\nD. uses an invertible group operation such as ``$\\oplus$\" for encryption.\n\nAnswer:", | |
| "Question: Using salt for UNIX passwords \\ldots?\n\nOptions:\nA. allows to speed up the verification for the server.\nB. makes the protocol secure against computationally unbounded adversary.\nC. allows to transmit the password over a non confidential channel without compromising the security.\nD. helps preventing dictionary attacks.\n\nAnswer:", | |
| "Question: Tick the \\emph{correct} assertion concerning WPA2?\n\nOptions:\nA. WPA2 uses RC4.\nB. WPA2 uses AES.\nC. WPA2 uses 64-bit keys.\nD. WPA2 is badly broken.\n\nAnswer:", | |
| "Question: Tick the \\textit{incorrect} assertion. In a typical (secure) TLS session \\ldots?\n\nOptions:\nA. the client authenticates himself to the server.\nB. the client and the server securely establish a symmetric key.\nC. the client first sends the list of accepted cipher suites.\nD. the client is sure that he is communicating with an authentic server.\n\nAnswer:", | |
| "Question: In User-Based Collaborative Filtering, which of the following is correct?\n\nOptions:\nA. Pearson Correlation Coefficient and Cosine Similarity have different value ranges, but return the same similarity ranking for the users\nB. Pearson Correlation Coefficient and Cosine Similarity have different value ranges and can return different similarity ranking for the users\nC. Pearson Correlation Coefficient and Cosine Similarity have the same value range, but can return different similarity ranking for the users\nD. Pearson Correlation Coefficient and Cosine Similarity have the same value range and return the same similarity ranking for the users.\n\nAnswer:", | |
| "Question: Which attribute gives the best split?A1PNa44b44A2PNx51y33A3PNt61j23?\n\nOptions:\nA. A1\nB. A3\nC. A2\nD. All the same\n\nAnswer:", | |
| "Question: Which of the following functions reaches a global maximum on the set $I$? (Note that $[.,.]$ and $(.,.)$ denote closed and open intervals respectively)\n $f_1(x) = -x^4$, $I = [-5,5]$\n\t\t $f_2(x) = \u0007rccos(x)$, $I = (-1,1)$\n\t\t $f_3(x) = x \\exp(-x)$, $I = (-\\infty, 0)$\n\t\t $f_4(x) = \\sin(\\cos(x))\\sin(x)$, $I= \\R_+$?\n\nOptions:\nA. $f_1, f_2, f_3, f_4\nB. $f_1, f_4$\nC. $f_1, f_3, f_4$\nD. $f_1, f_2, f_4$\n\nAnswer:", | |
| "Question: Suppose that you can prove the security of your symmetric encryption scheme against the following attacks. In which case is your scheme going to be the \\textbf{most} secure?\n\nOptions:\nA. Key recovery under known plaintext attack.\nB. Key recovery under chosen ciphertext attack.\nC. Decryption under known plaintext attack.\nD. Decryption under chosen ciphertext attack.\n\nAnswer:", | |
| "Question: Which of the following statements about code instrumentation is/are correct?\n\nOptions:\nA. We should instrument basic blocks when collecting edge coverage.\nB. change the original functionality.\nC. The instrumentation code for coverage collection should not\nD. Binary rewriting-based coverage collection has lower runtime\n\nAnswer:", | |
| "Question: In an x86 multiprocessor system with JOS, select all the correct options. Assume every Env has a single thread.\n\nOptions:\nA. Two Envs could run on the same processor simultaneously.\nB. Two Envs could run on two different processors simultaneously.\nC. One Env could run on two different processors simultaneously.\nD. One Env could run on two different processors at different times.\n\nAnswer:", | |
| "Question: The blockcipher AES \\ldots?\n\nOptions:\nA. is based on a Feistel scheme.\nB. can have a variable block size.\nC. should replace DES, as it is more secure than DES.\nD. has a round function that consists of the following operations: Expansion, AddBytes, ShiftRows, MixColumns.\n\nAnswer:", | |
| "Question: The number of non-zero entries in a column of a term-document matrix indicates:?\n\nOptions:\nA. how many terms of the vocabulary a document contains\nB. how often a term of the vocabulary occurs in a document\nC. how relevant a term is for a document\nD. none of the other responses is correct\n\nAnswer:", | |
| "Question: A major specificity of natural languages is that they are inherently implicit and ambiguous. How should this be taken into account in the NLP perspective?\n(penalty for wrong ticks)?\n\nOptions:\nA. by teaching humans to talk and write in a way that reduces implicitness and ambiguity\nB. by designing NLP algorithms and data structures able to efficiently cope with very ambiguous representations\nC. by interacting with human experts to formulate precise interpretation rules for linguistic entities\nD. by increasing the amount of a priori knowledge that NLP systems are able to exploit\n\nAnswer:", | |
| "Question: A shadow stack...\n\nOptions:\nA. is used by exception handling for stack trace generation.\nB. needs to be protected against or hidden from arbitrary writes.\nC. is an exact copy of the original stack allowing for a complete integrity verification.\nD. offers stronger protection than stack canaries.\n\nAnswer:", | |
| "Question: Which of the following attack vectors apply to mobile Android systems?\n\nOptions:\nA. money and not in providing software updates, resulting in outdated\nB. Hardware vendors like \\$am\\$ung are primarily interested in making\nC. Apps may maliciously declare intent filters to receive intents\nD. software that is vulnerable to attacks.\n\nAnswer:", | |
| "Question: Tick the \\textbf{true} assertion. Let $X$ be a random variable defined by the visible face showing up when throwing a dice. Its expected value $E(X)$ is:?\n\nOptions:\nA. 3.5\nB. 3\nC. 1\nD. 4\n\nAnswer:", | |
| "Question: Which of the following is/are true about fuzzing libraries?\n\nOptions:\nA. fuzzed independently.\nB. To fuzz effectively, every argument of every function must be\nC. single ``binary'' is available for a complex API.\nD. Fuzzing libraries is harder than standalone executables as no\n\nAnswer:", | |
| "Question: Select all sentences that are true.You will get a penalty for wrong ticks.\n\nOptions:\nA. PoS tagging becomes easy if lemmatization has already been performed on the words\nB. Lemmatization reduces word forms into their surface forms\nC. PoS tagging aims to resolve morpho-syntactic ambiguities associated with words\nD. PoS tagging aims to suppress lexical variability of words\n\nAnswer:", | |
| "Question: Let $E$ be a finite ground set and let $\\mathcal{I}$ be a family of ground sets. Which of the following definitions of $\\mathcal{I}$ guarantees that $M = (E, \\mathcal{I})$ is a matroid? \\begin{enumerate} \\item $E$ is the edges of an undirected bipartite graph and $\\mathcal{I} = \\{X \\subseteq E : \\mbox{$X$ is an acyclic edge set}\\}$. \\item $E$ is the edges of an undirected graph and $\\mathcal{I} = \\{X \\subseteq E : \\mbox{$X$ is an acyclic edge set}\\}$. \\item $E$ is the edges of an undirected bipartite graph and $\\mathcal{I} = \\{X \\subseteq E : \\mbox{$X$ is a matching}\\}$. \\item $E$ is the edges of an undirected graph and $\\mathcal{I} = \\{X \\subseteq E : \\mbox{$X$ is a matching}\\}$. \\item $E = \\{1, 2, \\ldots, n\\}$ is the set of indices of vectors $v_1, \\ldots, v_n \\in \\mathbb{R}^d$ and \\\\$\\mathcal{I} = \\{X \\subseteq E : \\mbox{the vectors $\\{v_i : i \\in X\\}$ are linearly \\emph{dependent}}\\}$. \\item $E = \\{1, 2, \\ldots, n\\}$ is the set of indices of vectors $v_1, \\ldots, v_n \\in \\mathbb{R}^d$ and \\\\$\\mathcal{I} = \\{X \\subseteq E : \\mbox{the vectors $\\{v_i : i \\in X\\}$ are linearly \\emph{independent}}\\}$. \\end{enumerate} The definitions of $\\mathcal{I}$ that guarantees that $M = (E, \\mathcal{I})$ is a matroid are:?\n\nOptions:\nA. (a), (b), (c), (d), (f)\nB. (a), (c), (d), (e)\nC. (a), (b), (c), (f)\nD. (a), (c), (f)\n\nAnswer:", | |
| "Question: Consider a hash function $H$ with $n$ output bits. Tick the \\emph{incorrect} assertion.\n\nOptions:\nA. Due to birthday paradox, an output collision of $H$ can be found much faster than with running time $2^n$.\nB. It is possible to find an output collision of $H$ with $O(2^{\\frac{n}{2}})$ memory and $O(2^{\\frac{n}{2}})$ running time.\nC. It is possible to find an output collision of $H$ with $O(1)$ memory and $O(2^{\\frac{n}{2}})$ running time.\nD. It is possible to find an output collision of $H$ with $O(2^{\\frac{n}{2}})$ memory and $O(1)$ running time.\n\nAnswer:", | |
| "Question: Suppose traditional inode pointer structure in ext3, i.e. 12 direct pointers, 1 singly, 1 doubly and 1 triply indirect pointer. Further suppose block size of 1kB and a 64-bit pointer. What is the maximal possible size for a single file?\n\nOptions:\nA. 4MB\nB. 1GB\nC. 512kB\nD. 2MB\n\nAnswer:", | |
| "Question: Which of the following hold(s) true about update deployment in the secure\n development lifecycle?\n\nOptions:\nA. Not allowing rolling back to previous versions is necessary\nB. You should always deploy third party updates automatically\nC. and immediately in your project.\nD. Updates may bring new code that may be buggy, so additional\n\nAnswer:", | |
| "Question: For each of the sub-questions of this question (next page), tick/check the corresponding box if the presented sentence is correct\nat the corresponding level (for a human). There will be a penalty for wrong boxes ticked/checked.The mouse lost a feather as it took off.\n\nOptions:\nA. syntactic\nB. lexical\nC. pragmatic\nD. semantic\n\nAnswer:", | |
| "Question: Which one of the following notions means that ``the information must be protected against any malicious modification''?\n\nOptions:\nA. privacy.\nB. integrity.\nC. confidentiality.\nD. reliability.\n\nAnswer:", | |
| "Question: Under which condition is an element $x\\in \\mathbb{Z}_n$ invertible?\n\nOptions:\nA. $\\mathsf{gcd}(x,\\varphi (n)) = 1$.\nB. $\\mathsf{gcd}(x,n-1) = 1$.\nC. $\\mathsf{gcd}(x,n) = 1$.\nD. $\\mathsf{gcd}(x,n) \\ne 1$.\n\nAnswer:", | |
| "Question: Choose the \\emph{correct} statement.\n\nOptions:\nA. $\\mathbb{Z}_n$ is a field $\\Leftrightarrow$ $n$ is a composite number\nB. $\\mathbb{Z}_n$ is a field $\\Leftrightarrow$ $\\mathbb{Z}_n^* = \\mathbb{Z}_n$\nC. $\\mathbb{Z}_n$ is a field $\\Leftrightarrow$ $n$ is a prime\nD. $\\mathbb{Z}_n$ is a field $\\Leftrightarrow$ $\\mathbb{Z}_n^* = \\emptyset$\n\nAnswer:", | |
| "Question: What is the value of \\(f(4)\\) where \\(f\\) is defined as \\(f(0) = f(1) = 1\\) and \\(f(n) = 2f(n - 1) + 3f(n - 2)\\) for integers \\(n \\geq 2\\)?\n\nOptions:\nA. 41\nB. 45\nC. 39\nD. 43\n\nAnswer:", | |
| "Question: Tick the non-commutative group.\n\nOptions:\nA. $\\mathbb{Z}_n^*$.\nB. $S_n$, the set of all permutations over the set $\\left\\{1,2,\\dots,n \\right\\}$.\nC. $E_{a,b}(K)$, an elliptic curve over a field $K$.\nD. $\\mathsf{GF}(2^k)^* $ .\n\nAnswer:", | |
| "Question: Select which statements are true about the CYK algorithm.A penalty will be applied for any incorrect answers.\n\nOptions:\nA. The Context-Free Grammar used with the CYK algorithm has to be converted into extended Chomsky normal form.\nB. It is a top-down chart parsing algorithm.\nC. Its time complexity is \\( O(n^3) \\), where\u00a0\\( n \\) is the length of sequence of words to be parsed.\nD. Its time complexity decreases when the grammar is regular.\n\nAnswer:", | |
| "Question: Which one of the following notions means that ``the information should make clear who the author of it is''?\n\nOptions:\nA. authentication\nB. steganograhy\nC. privacy\nD. confidentiality\n\nAnswer:", | |
| "Question: DES uses a key of size\\dots?\n\nOptions:\nA. $48$ bits\nB. $56$ bits\nC. $64$ bits\nD. $128$ bits\n\nAnswer:", | |
| "Question: To obtain a security of $2^{80}$ in a hash function against collisions one needs a hash output of size?\n\nOptions:\nA. $80$ bits.\nB. $40$ bits.\nC. $120$ bits.\nD. $160$ bits.\n\nAnswer:", | |
| "Question: Thang, Jeremie and Tugrulcan have built their own search engines. For a query Q, they got precision scores of 0.6, 0.7, 0.8 respectively. Their F1 scores (calculated by same parameters) are same. Whose search engine has a higher recall on Q?\n\nOptions:\nA. Thang\nB. Jeremie\nC. Tugrulcan\nD. We need more information\n\nAnswer:", | |
| "Question: A passive adversary can \\ldots?\n\nOptions:\nA. do nothing.\nB. only listen to communications.\nC. only interfere with client or server communications.\nD. only replace some communication messages by others.\n\nAnswer:", | |
| "Question: Tick the \\emph{false} assertion. A hash function can have the following roles:?\n\nOptions:\nA. Domain extender.\nB. Block cipher.\nC. Commitment.\nD. Pseudorandom generator.\n\nAnswer:", | |
| "Question: Tick the \\textbf{incorrect} assertion. Using sequential composition,\\dots?\n\nOptions:\nA. one can make a protocol more Zero-Knowledge.\nB. one can improve the soundness of a protocol.\nC. one can improve the completeness of a protocol.\nD. one has to repeat a protocol a given number of times using fresh random coins.\n\nAnswer:", | |
| "Question: Let $H$ be a hash function based on the Merkle-Damg{\\aa}rd construction. The Merkle-Damg{\\aa}rd theorem says that \\dots?\n\nOptions:\nA. \\dots $H$ is collision-resistant when the compression function is collision-resistant.\nB. \\dots the compression function is collision-resistant when $H$ is collision-resistant.\nC. \\dots $H$ is collision-resistant.\nD. \\dots $H$ is not collision-resistant.\n\nAnswer:", | |
| "Question: Given the distribution $P_0$ of a normal coin, i.e. $P_0(0)=P_0(1)=\\frac{1}{2}$, and distribution $P_1$ of a biased coin, where $P_1(0)=\\frac{1}{3}$ and $P_1(1) = \\frac{2}{3}$ , the maximal advantage of a distinguisher using a single sample is\\dots?\n\nOptions:\nA. $\\frac{1}{6}$.\nB. $3$.\nC. $\\frac{1}{3}$.\nD. $0$.\n\nAnswer:", | |
| "Question: The needed number of DES operations to attack DES is about ...\n\nOptions:\nA. $2^{54}.$\nB. $2^{64}.$\nC. $2^{128}.$\nD. $2^{256}.$\n\nAnswer:", | |
| "Question: Tick the \\textbf{false} statement. Enigma ...\n\nOptions:\nA. was broken mainly because of design flaws in the patented documents.\nB. was used by the German armies in World War 2.\nC. relies on Kerchkoffs's Principle.\nD. could be plugged into a radio transmitter.\n\nAnswer:", | |
| "Question: Due to the birthday paradox, a collision search in a hash function with $n$-bit output has complexity\\dots?\n\nOptions:\nA. $2^{\\sqrt{n}}$\nB. $\\sqrt{2^n}$\nC. $2^n$\nD. $2^{n-1}$\n\nAnswer:", | |
| "Question: Which is NOT a mobile telephony protocol?\n\nOptions:\nA. WPA\nB. DECT\nC. EDGE\nD. CDMA\n\nAnswer:", | |
| "Question: Moore's Law ...\n\nOptions:\nA. is an empirical law.\nB. says that the cost of computers doubles every 18 months.\nC. will allow to break AES in 2015.\nD. is a main reason for discarding MD5 hash function.\n\nAnswer:", | |
| "Question: Consider the language $L_{\\text{best}}=\\{\\text{``Advanced crypto is great!''}^i \\mid i \\in \\mathbb{N}\\}$. Tick an assertion.\n\nOptions:\nA. $L_{\\text{best}} \\in \\mathcal{P}$.\nB. Every $x\\in L_{\\text{best}}$ has a logical value equal to $\\mathsf{True}$. The bigger $|x|$, the more true it is.\nC. Only Chuck Norris can recognize $L_{\\text{best}}$ in constant time. But he can recognize any language in constant time.\nD. $L_{\\text{best}}$ is a very very good language. We have never seen any more bester language. It's huge.\n\nAnswer:", | |
| "Question: Does the following code snippet contain bugs? If so, which line(s)?\n\nOptions:\nA. Line 3 has a bug / vuln\nB. There is no bug / vuln in the code snippet\nC. Line 1 has a bug / vuln\nD. Line 4 has a bug / vuln\n\nAnswer:", | |
| "Question: Does the following code snippet contain bugs? If so, which line(s)?\n\nOptions:\nA. Line 1 has a bug / vuln\nB. Line 4 has a bug / vuln\nC. Line 2 has a bug / vuln\nD. There is no bug / vuln in the code snippet\n\nAnswer:", | |
| "Question: What is the difference between C++'s \\texttt{static\\_cast} and \\texttt{dynamic\\_cast}?\n\nOptions:\nA. \\texttt{static\\_cast} does not perform any kind of runtime check, while \\texttt{dynamic\\_cast} performs runtime checks on the validity of the cast.\nB. \\texttt{static\\_cast} is faster but less safe than \\texttt{dynamic\\_cast}.\nC. \\texttt{static\\_cast} can only be applied to static classes whereas \\texttt{dynamic\\_cast} works for any class.\nD. \\texttt{static\\_cast} does not work on already-casted objects, while \\texttt{dynamic\\_cast} works always.\n\nAnswer:", | |
| "Question: Given that JOS has correctly initialized the IDT and installed all the interrupt handlers. Which of the following will JOS do if the CPU with CPL = 3 tries to read the memory in data segment with DPL = 0?\n\nOptions:\nA. Calling the Page Fault Handler.\nB. Calling the General Protection Fault handler.\nC. Shuting down the system .\nD. Reading out the memory content successfully.\n\nAnswer:", | |
| "Question: What are possible morphological analyses of \"drinks\"?(Penalty for wrong ticks)?\n\nOptions:\nA. drink+VERB+p\nB. drink+ADJ\nC. drink+N+p\nD. N+s\n\nAnswer:", | |
| "Question: Which sets are countable (Multiple answers can be correct) :?\n\nOptions:\nA. The set of string of finite length of first names starting with the letter P\nB. The set of real numbers containing at least 100 3's in their decimal representation\nC. The set of natural numbers containing at least one 3 in their decimal representation\nD. \\(U-C\\) with \\(U\\) an uncountable set and \\(C\\) a countable set\n\nAnswer:", | |
| "Question: Let S(x) be the statement \u201cx has been in a lake\u201d and L(x) be the statement \u201cx lives in Lausanne\u201d and the domain of x consists of all the humans in the world.\n\nOptions:\nA. S(x) \\wedge L(x)) \\wedge \forall y \\left[ \nB. \\( \\exists x \\Bigr[ (\nC. S(y) \\wedge \nD. L(y)) \\wedge (x=y) \right] \\Bigr] \\)\n\nAnswer:", | |
| "Question: ``In any finite group, the order of an element is a factor of the order of the group\" is stated by?\n\nOptions:\nA. the Lagrange Theorem.\nB. the Chinese Remainder Theorem.\nC. the Fermat's Little Theorem.\nD. the Fundamental Theorem of Calculus.\n\nAnswer:", | |
| "Question: Let $X$ and $K$ be two independent random variables in a group $G$ with $n$ elements and where $K$ is uniformly distributed over $G$. Let $Y = X+K$. Then, for any $y\\in G$, $\\Pr[Y=y]=$\\dots?\n\nOptions:\nA. $1/n$.\nB. $\\Pr[X=y]$.\nC. $1/n^2$.\nD. $1/(n-1)$.\n\nAnswer:", | |
| "Question: Select the \\emph{incorrect} statement. Bluetooth is?\n\nOptions:\nA. a short-range wireless technology.\nB. designed both for data and voice transmission.\nC. a standard for RFID tags.\nD. able to transmit 1Mbit/sec in 10m distance.\n\nAnswer:", | |
| "Question: Which of the following is the powerset of \\(A= \\{\\{\\emptyset\\}\\}\\):?\n\nOptions:\nA. \\( \\emptyset \\)\nB. \\( \\{\\emptyset\\} \\)\nC. \\( \\{\\{\\emptyset\\}, \\emptyset\\} \\)\nD. \\( \\{\\{\\{\\emptyset\\}\\}, \\emptyset\\} \\)\n\nAnswer:", | |
| "Question: Standard encryption threats do not include:?\n\nOptions:\nA. Known-plaintext attacks.\nB. Chosen-plaintext attacks.\nC. Universal forgeries.\nD. Key-recovery attacks.\n\nAnswer:", | |
| "Question: What could Out of Vocabulary (OoV) forms consist of?\u00a0Select all that apply.A penalty will be applied for wrong answers.\n\nOptions:\nA. Neologisms\nB. Words with spelling errors\nC. Words borrowed from other languages\nD. Words from the lexicon\n\nAnswer:", | |
| "Question: Let \\(A = \\{a, b, c, d, ..., z\\}\\) be the set of lower cased English letters.\n\nLet \\(S = \\{a, b, ab, cd, ae, xy, ord, ...\\}\\) be the set of all strings using \\(A\\) as an alphabet.\n\nGiven \\(s\\in S\\), \\(N(s)\\) is the number of vowels in \\(s\\).\n\nFor example,\\(N(algrzqi) = 2\\), \\(N(bebebe) = 3\\).\n\nWe say \\((s, t)\\) belongs to relation \\(R\\) if \\(N(s) \\leq N(t)\\).\n\nWhich of the following statements are true (more than one answer can be correct) ?\n\nOptions:\nA. \\(R\\) is reflexive. \nB. \\(R\\) is transitive.\nC. \\(R\\) is symmetric.\nD. \\(R\\) is not an equivalence relation.\n\nAnswer:", | |
| "Question: The group $\\mathbb{Z}_{60}^*$ has \\ldots?\n\nOptions:\nA. 16 elements.\nB. 60 elements.\nC. 59 elements.\nD. 32 elements.\n\nAnswer:", | |
| "Question: Which of the following statements are true about DEP?\n\nOptions:\nA. DEP prohibits code injection but not code reuse.\nB. A single bit is used to indicate that a memory region is DEP protected.\nC. Even if it is called Enhanced Virus Protection on AMD hardware, it does not protect against viruses.\nD. Data Execution Prevention is usually handled at byte-level granularity.\n\nAnswer:", | |
| "Question: Consider a Generative Adversarial Network (GAN) which successfully produces images of goats. Which of the following statements is false?\n ?\n\nOptions:\nA. The discriminator can be used to classify images as goat vs non-goat.\nB. The generator aims to learn the distribution of goat images.\nC. After the training, the discriminator loss should ideally reach a constant value.\nD. The generator can produce unseen images of goats.\n\nAnswer:", | |
| "Question: In general, what is true regarding Fagin's algorithm?\n\nOptions:\nA. It performs a complete scan over the posting files\nB. It provably returns the k documents with the largest aggregate scores\nC. Posting files need to be indexed by the TF-IDF weights\nD. It never reads more than (kn)\u00bd entries from a posting list\n\nAnswer:", | |
| "Question: When applying Fagin\u2019s algorithm for a query with three different terms for finding the k top documents, the algorithm will scan?\n\nOptions:\nA. 2 different lists\nB. 3 different lists\nC. k different lists\nD. t depends how many rounds are taken\n\nAnswer:", | |
| "Question: In a Ranked Retrieval result, the result at position k is non-relevant and at k+1 is relevant. Which of the following is always true (P@k and R@k are the precision and recall of the result set consisting of the k top ranked documents)?\n\nOptions:\nA. P@k-1 > P@k+1\nB. P@k-1 = P@k+1\nC. R@k-1 < R@k+1\nD. R@k-1 = R@k+1\n\nAnswer:", | |
| "Question: Consider a binary classification problem with classifier $f(\\mathbf{x})$ given by $$ f(\\mathbf{x})= \\begin{cases}1, & g(\\mathbf{x}) \\geq 0 \\\\ -1, & g(\\mathbf{x})<0\\end{cases} $$ and $\\mathbf{x} \\in \\mathbb{R}^{6}$. Consider a specific pair $(\\mathbf{x}, y=1)$ and assume that $g(\\mathbf{x})=8$. In particular this means that this point is classified correctly by $f$. Assume further that we have computed the gradient of $g$ at $\\mathbf{x}$ to be $\\nabla_{\\mathbf{x}} g(\\mathbf{x})=(+1,-2,+3,-4,+5,-6)$. You are allowed to make one step in order to (hopefully) find an adversarial example. In the following four questions, assume $\\epsilon=1$. Which offset $\\delta$ with $\\|\\delta\\|_{1} \\leq 1$ yields the smallest value for $g(\\mathbf{x}+\\delta)$, assuming that $g$ is (locally) linear?\n\nOptions:\nA. $(0,0,0,0,0,1)$\nB. $(+1,-2,+3,-4,+5,-6)$\nC. $(-1,+2,-3,+4,-5,+6)$\nD. $(-1,+1,-1,+1,-1,+1)$\n\nAnswer:", | |
| "Question: Consider a linear model $\\hat{y} = xv ^\top \\wv$ with the squared loss under an $\\ell_\\infty$-bounded adversarial perturbation. For a single point $(xv, y)$, it corresponds to the following objective:\n \begin{align}\n \\max_{\tilde{xv}:\\ \\|xv-\tilde{xv}\\|_\\infty\\leq \\epsilon} \\left(y - \tilde{xv} ^\top \\wv \right)^{2},\n \tag{OP}\\AMClabel{eq:opt_adv_regression}\n \\end{align}\n where $\\|xv-\tilde{xv}\\|_\\infty\\leq \\epsilon$ denotes the $\\ell_\\infty$-norm, i.e. $|x_i - \tilde{x}_i| \\leq \u000barepsilon$ for every $i$.\n \\\n Assume that $\\wv = (3, -2)^\top$, $xv = (-1, 2)^\top$, $y=2$. What is the optimal $\tilde{xv}^\\star$ that maximizes the objective in Eq.~(\\AMCref{eq:opt_adv_regression})?\n\t ?\n\nOptions:\nA. Other\nB. $(-1+\u000barepsilon, 2)^\top$\nC. $(-1-\u000barepsilon, 2-\u000barepsilon)^\top$\nD. $(-1-\u000barepsilon, 2)^\top$\n\nAnswer:", | |
| "Question: Which of the following primitives \\textit{cannot} be instantiated with a cryptographic hash function?\n\nOptions:\nA. A pseudo-random number generator.\nB. A commitment scheme.\nC. A public key encryption scheme.\nD. A key-derivation function.\n\nAnswer:", | |
| "Question: Consider the linear mask $\\alpha := \\mathsf{0xf0}$ and the input $X := \\mathsf{0xe2}$. We have $\\alpha \\cdot X = $ \\dots?\n\nOptions:\nA. $\\mathsf{0}$\nB. $\\mathsf{1}$\nC. $\\mathsf{0xe0}$\nD. $\\mathsf{0xf2}$\n\nAnswer:", | |
| "Question: For his awesome research, Tugrulcan is going to use the PageRank with teleportation and HITS algorithm, not on a network of webpages but on the retweet network of Twitter! The retweet network is a directed graph, where nodes are users and an edge going out from a user A and to a user B means that \"User A retweeted User B\". Which one is false about a Twitter bot that retweeted other users frequently but got never retweeted by other users or by itself?\n\nOptions:\nA. It will have a non-zero hub value\nB. It will have an authority value of zero\nC. It will have a PageRank of zero\nD. Its authority value will be equal to the hub value of a user who never retweets other users\n\nAnswer:", | |
| "Question: Tick the \\textbf{true} assertion. Assume that $C$ is an arbitrary random permutation.\n\nOptions:\nA. $\\mathsf{BestAdv}_n(C,C^\\ast)=\\mathsf{Dec}^n_{\\left|\\left|\\left|\\cdot\\right|\\right|\\right|_a}(C)$\nB. $\\mathsf{BestAdv}_n^{n.a.}(C,C^\\ast)=\\frac{1}{2}\\mathsf{Dec}^n_{\\left|\\left|\\left|\\cdot\\right|\\right|\\right|_a}(C)$\nC. $E(\\mathsf{LP}^{C}(a,b)) < \\frac{1}{2}$\nD. $\\mathsf{BestAdv}_n^{n.a.}(C,C^\\ast)=\\frac{1}{2}\\mathsf{Dec}^n_{\\left|\\left|\\left|\\cdot\\right|\\right|\\right|_\\infty}(C)$\n\nAnswer:", | |
| "Question: Consider a public key cryptosystem. The channel used to transmit the public key has to be\\dots?\n\nOptions:\nA. \\dots encrypted.\nB. \\dots authenticated.\nC. \\dots confidential.\nD. \\dots authenticated and confidential.\n\nAnswer:", | |
| "Question: In the physical representation of an inverted file, the size of the index file is typically in the order of (where n is the number of documents):?\n\nOptions:\nA. O(log(n))\nB. O(sqrt(n))\nC. O(n)\nD. O(n2)\n\nAnswer:", | |
| "Question: Tick the \\emph{incorrect} assertion. In order to ensure message sequentiality \\ldots?\n\nOptions:\nA. the participants can acknowledge the receipt of every message.\nB. we can use TLS.\nC. the messages that are sent contain an authenticated sequence number.\nD. we make sure the participants finish with the same last message.\n\nAnswer:", | |
| "Question: Select which statements are true regarding SCFGs.A penalty will be applied for any incorrect answers.\n\nOptions:\nA. The inside-outside algorithm is used to derive the stochastic coefficients of a SCFG in a supervised manner.\nB. The\u00a0probabilities of lexical rules of a SCFG\u00a0correspond to\u00a0emission probabilities of HMMs\u00a0for PoS tagging.\nC. The probabilities of all parse trees in a SCFG sums up to 1.\nD. The sum over all the probabilities of the rules of a SCFG that share the\u00a0same left hand side\u00a0should sum up to 1.\n\nAnswer:", | |
| "Question: Which is an appropriate method for fighting skewed distributions of class labels in classification?\n\nOptions:\nA. Include an over-proportional number of samples from the larger class\nB. Use leave-one-out cross validation\nC. Construct the validation set such that the class label distribution approximately matches the global distribution of the class labels\nD. Generate artificial data points for the most frequent classes\n\nAnswer:", | |
| "Question: An HMM model would not be an appropriate approach to identify?\n\nOptions:\nA. Named Entities\nB. Part-of-Speech tags\nC. Concepts\nD. Word n-grams\n\nAnswer:", | |
| "Question: Let $p$ and $q$ be two distinct prime numbers and let $x \\in \\mathbf{Z}_{pq}^*$. Which of the following assertion is always true in $\\mathbf{Z}_{pq}^*$?\n\nOptions:\nA. $x^{p} = 1$\nB. $x^{q} = 1$\nC. $x^{pq} = 1$\nD. $x^{(p-1)(q-1)} = 1$\n\nAnswer:", | |
| "Question: The CRT states?\n\nOptions:\nA. $\\mathbb{Z}_{mn} \\equiv \\mathbb{Z}_{m} \\cup \\mathbb{Z}_{n}$\nB. $\\mathbb{Z}_{mn} \\equiv \\mathbb{Z}_{m} \\setminus \\mathbb{Z}_{n}$\nC. $\\mathbb{Z}_{mn} \\equiv \\mathbb{Z}_{m} \\cap \\mathbb{Z}_{n}$\nD. $\\mathbb{Z}_{mn} \\equiv \\mathbb{Z}_{m} \\times \\mathbb{Z}_{n}$\n\nAnswer:", | |
| "Question: Tick the \\emph{incorrect} statement:?\n\nOptions:\nA. Zero-knowledge interactive proofs are interactive proofs.\nB. Black-box zero-knowledge interactive proofs are zero-knowledge interactive proofs.\nC. $\\Sigma$-protocols are zero-knowledge interactive proofs.\nD. $\\Sigma$-protocols are interactive proofs of knowledge.\n\nAnswer:", | |
| "Question: The exponent $\\lambda(21)$ of $\\mathbb{Z}_{21}^*$ is ...\n\nOptions:\nA. 21.\nB. 12.\nC. 6.\nD. 3.\n\nAnswer:", | |
| "Question: Which of the following scheduler policies are preemptive?\n\nOptions:\nA. FIFO (First In, First Out)\nB. SJF (Shortest Job First)\nC. STCF (Shortest Time to Completion First)\nD. RR (Round Robin)\n\nAnswer:", | |
| "Question: Tick the \\textbf{incorrect} assertion regarding plain Rabin, i.e., Rabin without any redundancy.\n\nOptions:\nA. The Rabin Key Recovery Problem relies on the discrete logarithm problem.\nB. Plain Rabin suffers from a chosen ciphertext key recovery attack.\nC. The decryption of plain Rabin is ambiguous.\nD. The Rabin Decryption Problem is equivalent to the factoring problem.\n\nAnswer:", | |
| "Question: How many necessary and sufficient queries do we need to distinguish the distribution $P$ from the uniform distribution $U$ with significant advantage?\n\nOptions:\nA. $d(P,U)^{-1}$\nB. $\\infty$\nC. $\\Theta(SEI(P)^{-1})$\nD. $\\Theta(C(P,U))$\n\nAnswer:", | |
| "Question: Let $n$ be an integer. The extended Euclidean algorithm is typically used to\\dots?\n\nOptions:\nA. \\dots perform the addition of two integers in $\\mathbf{Z}_n^*$.\nB. \\dots compute the inverse of an element in $\\mathbf{Z}_n^*$.\nC. \\dots compute the square of an element of $\\mathbf{Z}_n^*$.\nD. \\dots compute the order of $\\mathbf{Z}_n^*$.\n\nAnswer:", | |
| "Question: Which of the following statements about symbolic execution is/are correct?\n\nOptions:\nA. in programs with simple logics but large input space.\nB. Symbolic execution requires actually running the target\nC. Symbolic execution can always accurately model a system's\nD. environment (e.g., system calls, file I/O, and network I/O).\n\nAnswer:", | |
| "Question: Select all statements that are true.A penalty will be applied for any wrong answers.\n\nOptions:\nA. Phrase-structure grammars better describe selectional constraints.\nB. Dependency grammars describe functional dependencies between words in a sequence.\nC. Phrase-structure grammars are relatively better suited for fixed-order languages than free-order languages.\nD. Any context-free grammar can be transformed into Chomsky-Normal form.\n\nAnswer:", | |
| "Question: Given the following data structure: enum IntSet: \\t case Empty \\t case NonEmpty(x: Int, l: IntSet, r: IntSet) And the following lemmas, holding for all x: Int, xs: List[Int], ys: List[Int], l: IntSet and r: IntSet: (SizeNil) nil.size === 0 (SizeCons) (x :: xs).size === xs.size + 1 (ConcatSize) (xs ++ ys).size === xs.size + ys.size (TreeSizeEmpty) Empty.treeSize === 0 (TreeSizeNonEmpty) NonEmpty(x, l, r).treeSize === l.treeSize + r.treeSize + 1 (ToListEmpty) Empty.toList === nil (ToListNonEmpty) NonEmpty(x, l, r).toList === l.toList ++ (x :: r.toList) Let us prove the following lemma for all s: IntSet: (ToListSize) s.toList.size === s.treeSize We prove it by induction on s. Base case: s is Empty. Therefore, we need to prove: Empty.toList.size === Empty.treeSize Starting from the left hand-side (Empty.toList.size), what exact sequence of lemmas should we apply to get the right hand-side (Empty.treeSize)?\n\nOptions:\nA. ToListEmpty, TreeSizeEmpty, SizeNil\nB. SizeNil, TreeSizeEmpty, ToListEmpty\nC. SizeNil, ToListEmpty, TreeSizeEmpty\nD. ToListEmpty, SizeNil, TreeSizeEmpty\n\nAnswer:", | |
| "Question: Consider the program below. Tick the correct answer. def fun(x: List[Int]) = if x.isEmpty then None else Some(x) val lists = List(List(1, 2, 3), List(), List(4, 5, 6)) for \\t l <- lists \\t v1 <- fun(l) \\t v2 <- fun(v1) yield v2?\n\nOptions:\nA. This program compiles and the last statement has type List[Option[List[Int]]].\nB. This program compiles and the last statement has type List[List[Int]].\nC. This program compiles and the last statement has type List[Int].\nD. This program does not compile.\n\nAnswer:", | |
| "Question: Which symmetric key primitive is used in WPA2 encryption?\n\nOptions:\nA. RC4 CBC Mode\nB. KASUMI ECB Mode\nC. MD5 OFB Mode\nD. AES CCM Mode\n\nAnswer:", | |
| "Question: Consider an Sbox $S:\\{0,1\\}^m \\rightarrow \\{0,1\\}^m$. We have that \\ldots?\n\nOptions:\nA. $\\mathsf{DP}^S(0,b)=1$ if and only if $S$ is a permutation.\nB. $\\sum_{b\\in \\{0,1\\}^m} \\mathsf{DP}^S(a,b)$ is even.\nC. $\\sum_{b\\in \\{0,1\\}^m \\backslash \\{0\\}} \\mathsf{DP}^S(0,b)= 0$\nD. $\\mathsf{DP}^S(0,b)=1$ if and only if $m$ is odd.\n\nAnswer:", | |
| "Question: Tick the \\textbf{incorrect} assumption. A language $L$ is in NP if\\dots?\n\nOptions:\nA. $x \\in L$ can be decided in polynomial time.\nB. $x \\in L$ can be decided in polynomial time given a witness $w$.\nC. $L$ is NP-hard.\nD. $L$ (Turing-)reduces to a language $L_2$ with $L_2$ in $P$, i.e., if there is a polynomial deterministic Turing machine which recognizes $L$ when plugged to an oracle recognizing $L_2$.\n\nAnswer:", | |
| "Question: Tick the \\emph{incorrect} assertion. For a cipher $C$, decorrelation theory says that \\ldots?\n\nOptions:\nA. A decorrelation $0$ of order $1$ means perfect secrecy when used once.\nB. $\\mathsf{BestAdv}_n(C,C^\\ast)=\\frac{1}{2}\\mathsf{Dec}^n_{\\left|\\left|\\cdot\\right|\\right|_a}(C)$.\nC. A decorrelation $0$ of order $1$ always protects against linear cryptanalysis.\nD. $\\mathsf{Dec}^n(C_1\\circ C_2) \\leq \\mathsf{Dec}^n(C_1) \\times \\mathsf{Dec}^n(C_2)$, for $C_1$ and $C_2$ two independent random permutations.\n\nAnswer:", | |
| "Question: Tick the \\textit{incorrect} assertion. Given an RSA public key $(e,N)$, we \\textit{can} efficiently compute square roots in $\\mathbb{Z}_{N}^*$ if we have an efficient algorithm that \\ldots?\n\nOptions:\nA. factors $N$.\nB. recovers the corresponding secret key $d$.\nC. computes $\\varphi(N)$, the order of $\\mathbb{Z}_N^*$.\nD. given a $y\\in \\mathbb{Z}_N$ computes an $x$ such that $x^e \\equiv y \\pmod{N}$.\n\nAnswer:", | |
| "Question: Consider the exhaustive search of a uniformly distributed key in a set of size $N$. What is the complexity of the exhaustive search with time memory trade-off? (We assume that memory access is constant.)?\n\nOptions:\nA. precomputation: $O(N^{2/3})$, memory: $O(N^{2/3})$, time: $O(N^{2/3})$.\nB. precomputation: $O(N)$, memory: $O(N^{2/3})$, time: $O(N^{2/3})$.\nC. precomputation: $O(1)$, memory: $O(N^{2/3})$, time: $O(N^{2/3})$.\nD. precomputation: $O(N)$, memory: $O(1)$, time: $O(N^{2/3})$.\n\nAnswer:", | |
| "Question: Which of the following statements about testing is/are correct?\n\nOptions:\nA. susceptible to state space explosion.\nB. of the provided concrete input.\nC. over-approximation of targets.\nD. Compared to dynamic analysis, static analysis is more\n\nAnswer:", | |
| "Question: In deep learning, which of these are hyper-parameters?\n\nOptions:\nA. The learning rate\nB. The weights $\\mathbf{W}^{[l]}$ and biases $\\mathbf{b}^{[l]}$\nC. The type of weight initialization\nD. The number of layers\n\nAnswer:", | |
| "Question: Which statement is correct?\n\nOptions:\nA. The Viterbi algorithm works because words are independent in a sentence\nB. The Viterbi algorithm works because it is applied to an HMM model that makes an independence assumption on the word dependencies in sentences\nC. The Viterbi algorithm works because it makes an independence assumption on the word dependencies in sentences\nD. The Viterbi algorithm works because it is applied to an HMM model that captures independence of words in a sentence\n\nAnswer:", | |
| "Question: Select the \\emph{incorrect} statement. The brute force attack \\dots?\n\nOptions:\nA. can be applicable after decades according to Moore's law.\nB. can break a cipher with a $128$-bit key on your PC today.\nC. has higher worst case complexity than average case complexity.\nD. refers to a way of getting the secret key, exhaustively.\n\nAnswer:", | |
| "Question: Consider the composite function $f(x)=g(h(x))$, where all functions are $\\mathbb{R}$ to $\\mathbb{R}$. Which of the following is the weakest condition that guarantees that $f(x)$ is convex?\n\nOptions:\nA. $g(x)$ is convex and $h(x)$ is increasing\nB. $g(x)$ is convex and $g(x)$ and $h(x)$ are increasing\nC. $h(x)$ is convex and $g(x)$ and $h(x)$ are increasing\nD. $g(x)$ is convex and $g(x)$ is increasing\n\nAnswer:", | |
| "Question: Select all the statements that are true.A penalty will be applied for any incorrect answers selected.\n\nOptions:\nA. Cosine similarity is independent of the length of the documents.\nB. The Luhn law states that if a set of words are ranked by the decreasing order of their frequencies, the high-ranked words are the best features for identifying the topics that occur in the document collection.\nC. The order of words are ignored in the bag-of-words model.\nD. High values of document frequency means that the word is not very discriminative.\n\nAnswer:", | |
| "Question: Which of AddressSanitizer (ASan), MemorySanitizer (MemSan),\n UndefinedBehaviorSanitizer (UBSan) or ThreadSanitizer (TSan) can detect\n bugs (if any) in the following code snippet?\n\nOptions:\nA. ASan\nB. There is at least one bug in the snippet, but none of the\nC. mentioned sanitizers can detect it.\nD. There are no bugs in the snippet.\n\nAnswer:", | |
| "Question: What is the gradient of $\\mathbf{x}^{\\top} \\mathbf{W} \\mathbf{x}$ with respect to $\\mathbf{x}$ (written as a vector)?\n\nOptions:\nA. (g) $\\mathbf{W} \\mathbf{W}^{\\top}$.\nB. (e) $\\mathbf{x x}^{\\top}$.\nC. (f) $\\mathbf{x}^{\\top} \\mathbf{x}$\nD. (c) $\\square\\left(\\mathbf{W}+\\mathbf{W}^{\\top}\\right) \\mathbf{x}$\n\nAnswer:", | |
| "Question: Select the \\textbf{most accurate} answer. For a hash function to be secure (\\textbf{but still efficient}) against collision attacks in 2015, the output length should be\\dots?\n\nOptions:\nA. around 80 bits.\nB. around 160 bits.\nC. around 512 bits.\nD. around 1024 bits.\n\nAnswer:", | |
| "Question: Which of the following statements about symbolic execution is/are correct?\n\nOptions:\nA. Symbolic execution can efficiently handle and solve constraints\nB. State space explosion is a common challenge for symbolic\nC. execution.\nD. in programs with simple logics but large input space.\n\nAnswer:", | |
| "Question: Which of the following statements are true about command injection?\n\nOptions:\nA. shell commands (e.g., \\texttt{system()}) with less privileged alternatives\nB. such as \\texttt{read\\_file()}.\nC. between data plane and control/code plane.\nD. To mitigate command injection, it is best to replace powerful\n\nAnswer:", | |
| "Question: When using linear regression, which techniques improve your result? (One or multiple answers)?\n\nOptions:\nA. linear regression does not allow polynomial features\nB. polynomial combination of features\nC. adding new features that are non-linear combination of existing features\nD. because the linear nature needs to be preserved, non-linear combination of features are not allowed\n\nAnswer:", | |
| "Question: Church booleans are a representation of booleans in the lambda calculus. The Church encoding of true and false are functions of two parameters: Church encoding of tru: t => f => t Church encoding of fls: t => f => f What should replace ??? so that the following function computes not(b and c)? b => c => b ??? (not b)?\n\nOptions:\nA. (not b)\nB. (not c)\nC. tru\nD. fls\n\nAnswer:", | |
| "Question: Which of the following statements are true about command injection?\n\nOptions:\nA. The root cause of command injection is the lack of distinction\nB. invoke internal shell commands.\nC. Command injection allows the attacker to launch new processes or\nD. between data plane and control/code plane.\n\nAnswer:", | |
| "Question: Tick the \\textbf{false} assertion. Assume that $C$ is a random permutation.\n\nOptions:\nA. BestAdv$_n(C,C^\\ast)=\\frac{1}{2}Dec^n_{\\left|\\left|\\left|\\cdot\\right|\\right|\\right|_a}(C)$\nB. BestAdv$_n^{n.a.}(C,C^\\ast)=\\frac{1}{2}Dec^n_{\\left|\\left|\\left|\\cdot\\right|\\right|\\right|_\\infty}(C)$\nC. $E(LP^{C}(a,b))\\leq 1$\nD. $Dec^n(C\\circ C)\\leq Dec^n(C)^2$.\n\nAnswer:", | |
| "Question: Your input is \"Distributed Information Systems\". Your model tries to predict \"Distributed\" and \"Systems\" by leveraging the fact that these words are in the neighborhood of \"Information\". This model can be:?\n\nOptions:\nA. Bag of Words\nB. Word Embeddings\nC. LDA\nD. kNN\n\nAnswer:", | |
| "Question: With Fagin\u2019s algorithm, once k documents have been identified that occur in all of the lists?\n\nOptions:\nA. These are the top-k documents\nB. The top-k documents are among the documents seen so far\nC. The search has to continue in round-robin till the top-k documents are identified\nD. Other documents have to be searched to complete the top-k list\n\nAnswer:", | |
| "Question: Let the Boolean query be represented by {(1, 0, -1), (0, -1, 1)} and the document by (1, 0, 1). The document ?\n\nOptions:\nA. matches the query because it matches the first query vector\nB. matches the query because it matches the second query vector\nC. does not match the query because it does not match the first query vector\nD. does not match the query because it does not match the second query vector\n\nAnswer:", | |
| "Question: Tick the \\textit{correct} assertion.\n\nOptions:\nA. In a finite field $K$, every element has exactly two square roots.\nB. In a finite field $K$, 1 has exactly one square roots and it is 1.\nC. The set of quadratic residues in $\\mathbb{Z}_n$ is a field.\nD. An element can have more than two square roots in $\\mathbb{Z}_n$.\n\nAnswer:", | |
| "Question: Let $f$ be any hash function with output domain size $N$.\n\nOptions:\nA. One can find a collision in $f$ in $\\mathcal{O}(\\sqrt{N})$ using almost no memory.\nB. One can find a preimage in $f$ in $\\mathcal{O}(\\sqrt{N})$ using $\\sqrt{N}$ memory.\nC. One can find a second preimage in $f$ in $\\mathcal{O}(\\sqrt{N})$ using $\\sqrt{N}$ memory.\nD. The best collision attack against $f$ runs in time $\\mathcal{O}(N)$.\n\nAnswer:", | |
| "Question: When using linear regression, what can cause numerical instabilities? (One or multiple answers)?\n\nOptions:\nA. learning rate too small\nB. degeneracies in the features\nC. too much training data\nD. too little training data\n\nAnswer:", | |
| "Question: Select \\emph{incorrect} statement. The exhaustive search?\n\nOptions:\nA. can be used to find a secret key of AES.\nB. is a brute force attack.\nC. is not applicable against perfectly secure cipher.\nD. runs in time polynomial in the length of the key.\n\nAnswer:", | |
| "Question: A vending machine dispensing books of stamps accepts only $1 bills, $2 bills and $5 bills.\n\nFind a recurrence relation for the number of ways to deposit $$n$$ dollars in the vending machine, where the order in which the coins and bills are deposited matters.\n\nOptions:\nA. $$a_{n} = a_{n-1} + a_{n-2} + a_{n-5}$$\nB. $$a_{n+1} = a_{n-1} + a_{n-2} + a_{n-5}$$\nC. $$a_{n+1} = a_{n-1} + 2a_{n-2} + 5a_{n-5}$$\nD. $$a_{n} = a_{n-1} + 2a_{n-2} + 5a_{n-5}$$\n\nAnswer:", | |
| "Question: Let $E$ be a finite ground set and let $\\mathcal{I}$ be a family of ground sets. Which of the following definitions of $\\mathcal{I}$ guarantees that $M = (E, \\mathcal{I})$ is a matroid? \\begin{enumerate} \\item $E$ is the edges of an undirected bipartite graph and $\\mathcal{I} = \\{X \\subseteq E : \\mbox{$X$ is an acyclic edge set}\\}$. \\item $E$ is the edges of an undirected graph and $\\mathcal{I} = \\{X \\subseteq E : \\mbox{$X$ is an acyclic edge set}\\}$. \\item $E$ is the edges of an undirected bipartite graph and $\\mathcal{I} = \\{X \\subseteq E : \\mbox{$X$ is a matching}\\}$. \\item $E$ is the edges of an undirected graph and $\\mathcal{I} = \\{X \\subseteq E : \\mbox{$X$ is a matching}\\}$. \\item $E = \\{1, 2, \\ldots, n\\}$ is the set of indices of vectors $v_1, \\ldots, v_n \\in \\mathbb{R}^d$ and \\\\$\\mathcal{I} = \\{X \\subseteq E : \\mbox{the vectors $\\{v_i : i \\in X\\}$ are linearly \\emph{dependent}}\\}$. \\item $E = \\{1, 2, \\ldots, n\\}$ is the set of indices of vectors $v_1, \\ldots, v_n \\in \\mathbb{R}^d$ and \\\\$\\mathcal{I} = \\{X \\subseteq E : \\mbox{the vectors $\\{v_i : i \\in X\\}$ are linearly \\emph{independent}}\\}$. \\end{enumerate} The definitions of $\\mathcal{I}$ that guarantees that $M = (E, \\mathcal{I})$ is a matroid are:?\n\nOptions:\nA. (a), (b), (c), (e)\nB. (a), (b), (f)\nC. (a), (f)\nD. (a), (c), (d), (e)\n\nAnswer:", | |
| "Question: Which of AddressSanitizer (ASan), MemorySanitizer (MemSan),\n UndefinedBehaviorSanitizer (UBSan) or ThreadSanitizer (TSan) can detect\n bugs (if any) in the following code snippet?\n\nOptions:\nA. TSan\nB. UBSan\nC. MemSan\nD. There are no bugs in the snippet.\n\nAnswer:", | |
| "Question: What are the complexities for the single-target dictionary attacks, when there are $N$ keys?\n\nOptions:\nA. Preprocessing: $N$, Memory: $N$, Time: 1\nB. Preprocessing: $N$, Memory: $1$, Time: $N$\nC. Preprocessing: $1$, Memory: $N$, Time: $N$\nD. Preprocessing: $0$, Memory: $1$, Time: $\\sqrt{N}$\n\nAnswer:", | |
| "Question: Tick the \\textit{incorrect} assertion. Consider a device that is running a software implementation of the PKCS\\#1v1.5 RSA cryptosystem.\n\nOptions:\nA. Analysing the power consumption of the device during decryption may be used to help recover the secret key.\nB. Inducing computational errors in the device during encryption may help recover the secret key.\nC. Issues related to message formatting may be used to help recover the secret key.\nD. Measuring the timing of the decryption computation may be used to help recover the secret key.\n\nAnswer:", | |
| "Question: Identify the correct statement.\n\nOptions:\nA. None of the other options are correct.\nB. After training, and when the size of the vocabulary is large, a Skip-gram model would have higher space requirements than a GloVe model. We assume both models have the same number of dimensions (features), vocabulary, and are trained on the same corpus.\nC. Language models can be trained using either a multi-class(number of classes equal to the vocabulary size) classifier or a binary classifier to generate text.\nD. Language Models are useless for classification tasks in Natural Language Processing as they are only suited for text generation.\n\nAnswer:", | |
| "Question: We want to return, from the two posting lists below, the top-2 documents matching a query using Fagin\u2019s algorithm with the aggregation function taken as the sum of the tf-idf weights. How many entries (total of both lists) are accessed in the first phase of the algorithm performing round robin starting at List 1 (i.e., before performing the random access)?List 1List 2documenttf-idfdocumenttf-idfd30.8d10.8d20.6d30.6d10.5d40.5d40.4d20.4?\n\nOptions:\nA. 2\nB. 4\nC. 6\nD. 8\n\nAnswer:", | |
| "Question: The CRT implies?\n\nOptions:\nA. $\\phi(mn)=\\phi(m)\\phi(n)$, for $gcd(m,n)=1$.\nB. $\\phi(mn)=\\phi(m)+\\phi(n)$, for $gcd(m,n)=1$.\nC. $\\phi(mn)=\\phi(m)^n$, for $m\\ne n$.\nD. $\\phi(mn)=\\phi(m)-n$, for $m > n$.\n\nAnswer:", | |
| "Question: Which of the following is false regarding K-means and DBSCAN?\n\nOptions:\nA. K-means does not handle outliers, while DBSCAN does\nB. K-means takes the number of clusters as parameter, while DBSCAN does not take any parameter\nC. K-means does many iterations, while DBSCAN does not\nD. Both are unsupervised\n\nAnswer:", | |
| "Question: Which mode of operation is similar to a stream cipher?\n\nOptions:\nA. ECB\nB. OFB\nC. CFB\nD. CBC\n\nAnswer:", | |
| "Question: Which of the following hold true for cross-site scripting (XSS)?\n\nOptions:\nA. XSS is a form of code injection that gives the attacker\nB. arbitrary code execution.\nC. Reflected XSS requires that the server stores the injected\nD. code but the user does not need to click on any special link.\n\nAnswer:", | |
| "Question: Which one of these is \\emph{not} a skeleton used to design block ciphers?\n\nOptions:\nA. a Feistel scheme\nB. a Subsitution-permutation network\nC. a Merkle-Damg{\\aa}rd construction\nD. a Lay-Massey Scheme\n\nAnswer:", | |
| "Question: Which of the following statements is \textbf{incorrect} ? Training a model with $L_1$-regularization ...\n\nOptions:\nA. can reduce the storage cost of the final model.\nB. is used to help escaping local minima during training.\nC. can reduce overfitting.\nD. can be named Lasso regression when in combination with an MSE loss function and a linear model.\n\nAnswer:", | |
| "Question: Given two distributions $P_0$ and $P_1$ over a discrete set $Z$, the maximal advantage of a distinguisher using a single sample is\\dots?\n\nOptions:\nA. $\\frac{1}{2}\\sum_{x\\in Z} | P_0(x) - P_1(x)|$.\nB. $1 - \\prod_{x\\in Z}(P_0(x)-P_1(x))^2$.\nC. $1$ if $P_0 = P_1$ and $0$ otherwise.\nD. always $1$ for computationally unbounded adversaries.\n\nAnswer:", | |
| "Question: Tick the \\emph{incorrect} assertion regarding the Diffie-Hellman key exchange?\n\nOptions:\nA. It is secure under a passive adversary.\nB. It is secure when working in a subgroup of $\\mathbb{Z}_p^*$ of big prime order.\nC. It is secure under an active Man-In-The-Middle attack.\nD. It is not secure when working in $\\mathbb{Z}_n$ with $n$ having only small prime factors.\n\nAnswer:", | |
| "Question: Let $n$ be a positive integer. The Fermat test most likely outputs ``prime'' \\dots?\n\nOptions:\nA. only when $n$ is prime.\nB. only when $n$ is non-prime.\nC. when $n$ is prime or when $n$ is not a Carmichael number.\nD. when $n$ is prime or when $n$ is a Carmichael number.\n\nAnswer:", | |
| "Question: Which is true?\n\nOptions:\nA. Hand-written patterns are in general more precise than classifiers\nB. Hand-written patterns cannot exploit syntactic features\nC. Supervised classifiers do not require any human input\nD. Supervised classifiers can only detect typed statements\n\nAnswer:", | |
| "Question: Heidi is working on some linear regression problem to predict the price of goat milk. When training her model, she gets a loss of 0. Which of the statements below \\textbf{must then be true}?\n\nOptions:\nA. We must have $y^{(i)} = 0 \\ \\ \\forall \\ i \\in \\{1, ..., N\\}$\nB. The weights $\\mathbf{w}$ must all be 0 so that $\\hat{y}^{(i)} = \\mathbf{w}^T \\boldsymbol{x}^{(i)} = 0.$\nC. Our training set can be fit perfectly by a hyperplane (e.g., fit perfectly by a straight line if our data is 2-dimensional).\nD. Gradient descent is stuck at a local minima and fails to find the true global minimum.\n\nAnswer:", | |
| "Question: Tick the \\emph{true} assertion.\n\nOptions:\nA. In an universal forgery the adversary has stronger objectives than in an existential forgery.\nB. In a MAC forgery under known message attack the adversary is able to request the authentication of several messages.\nC. In an existential forgery the adversary is able to forge a valid MAC for an arbitrary message.\nD. A MAC provides authentication, integrity and confidentiality.\n\nAnswer:", | |
| "Question: How does LSI querying work?\n\nOptions:\nA. The query vector is treated as an additional term; then cosine similarity is computed\nB. The query vector is transformed by Matrix S; then cosine similarity is computed\nC. The query vector is treated as an additional document; then cosine similarity is computed\nD. The query vector is multiplied with an orthonormal matrix; then cosine similarity is computed\n\nAnswer:", | |
| "Question: Consider a regression model where data $(x,y)$ is generated by input $x$ uniformly randomly sampled from $[0,1]$ and $y(x) = x^2 + \\epsilon$,\n\t where $\\epsilon$ is random noise with mean 0 and variance 1.\n Two models are carried out for regression:\n model A is a trained quadratic function $g(x; \\wv) = w_2 x^2 + w_1 x + w_0$ where $\\wv = (w_0, w_1, w_2)^\top\\in\\mathbb R^3$,\n and model B is a constant function $h(x) = 1/2$.\n Then compared to model B, model A has ?\n\nOptions:\nA. higher bias, higher variance.\nB. higher bias, lower variance.\nC. lower bias, higher variance.\nD. lower bias, lower variance.\n\nAnswer:", | |
| "Question: The following member of the exponential family represents a scalar Gaussian: $p(y)=\\exp \\left\\{(2,-1)\\left(y, y^{2}\\right)^{\\top}-\\right.$ $\\left.1-\\frac{1}{2} \\ln (\\pi)\\right\\}$. What are the mean $\\mu$ and the variance $\\sigma^{2}$ ?\n\nOptions:\nA. (f) $\\mu=1, \\sigma^{2}=\\frac{1}{2}$.\nB. (e) $\\mu=0, \\sigma^{2}=\\frac{1}{2}$.\nC. (g) $\\mu=-1, \\sigma^{2}=1$.\nD. (b) $\\mu=0, \\sigma^{2}=0$.\n\nAnswer:", | |
| "Question: What is the mean squared error of $f$ for a sample, where $\\textbf{x}$ is an input, $y$ a target and $f(\\textbf{x},W)$ the mapping function ? \n\t(One answer)?\n\nOptions:\nA. $||y - f(\\textbf{x},W)||^2 $ \nB. $||y - f(\\textbf{x},W)|| $\nC. $-\\log(P(y=i | \\textbf{x})) = -\\log(\\frac{e^{\\textbf{f}_i(\\textbf{x},W)}}{\\sum_j e^{\\textbf{f}_j(\\textbf{x},W)}})$ \nD. $P(y=i |\\textbf{x}) = \\frac{e^{\\textbf{f}_i(\\textbf{x},W)}}{\\sum_j e^{\\textbf{f}_j(\\textbf{x},W)}}$ \n\nAnswer:", | |
| "Question: Which operation does the following function implement? extension [U](l: List[U]) def secret[T](t: T)(f: (T, U) => T): T = \\t var res = t \\t l.foreach(u => res = f(res, u)) \\t res?\n\nOptions:\nA. reduceLeft\nB. reduceRight\nC. foldLeft\nD. foldRight\n\nAnswer:", | |
| "Question: Which protocol does \\emph{not} use RC4.\n\nOptions:\nA. WEP\nB. WPA\nC. WPA2\nD. TLS\n\nAnswer:", | |
| "Question: For two $\\ell$-bit numbers, the \\textbf{tightest} complexity to compute their multiplication is?\n\nOptions:\nA. $\\mathcal{O}(\\ell)$.\nB. $\\mathcal{O}(\\sqrt{\\ell})$.\nC. $\\mathcal{O}(\\ell^2)$.\nD. $\\mathcal{O}(\\ell^2 \\log{\\ell})$.\n\nAnswer:", | |
| "Question: The multiplication of two quadratic residues ...\n\nOptions:\nA. is not a quadratic residue.\nB. is always a quadratic residue.\nC. can be either a quadratic residue or not.\nD. is non-invertible.\n\nAnswer:", | |
| "Question: Which of the following statements is correct?\n\nOptions:\nA. When applying stochastic gradient descent on the objective function $f(\\boldsymbol{w}):=\\sum_{n=1}^{30}\\left\\|\\boldsymbol{w}-\\boldsymbol{x}_{n}\\right\\|^{2}$ where $\\boldsymbol{x}_{n}$ are the datapoints, a stochastic gradient step is roughly $30 \\times$ faster than a full gradient step.\nB. In practice, it could be good to let your model first overfit your task, and then apply drop-out or other regularization techniques.\nC. When applying stochastic gradient descent on the objective function $f(\\boldsymbol{w}):=\\sum_{n=1}^{30} n \\cdot\\|\\boldsymbol{w}\\|^{2}$, a stochastic gradient (for the $n$-th summand) is given by $2 n \\cdot \\boldsymbol{w}$.\nD. The function $f(\\boldsymbol{u} ; \\boldsymbol{v}):=g\\left(\\boldsymbol{u} \\boldsymbol{v}^{\\top}\\right)$ is convex over the set of pairs of vectors $(\\boldsymbol{u} ; \\boldsymbol{v}) \\in \\mathbb{R}^{2} \\times \\mathbb{R}^{2}$, when $g: \\mathbb{R}^{2 \\times 2} \\rightarrow \\mathbb{R}$ is defined as $g(\\boldsymbol{X}):=X_{12}+X_{21}$.\n\nAnswer:", | |
| "Question: For this question,\u00a0one or more\u00a0assertions can be correct. Tick only the correct assertion(s).\u00a0There will be a penalty for wrong assertions ticked.Which of the following associations can be considered as illustrative examples for inflectional\nmorphology (with here the simplifying assumption that canonical forms are restricted to the roots\nonly)?\n\nOptions:\nA. (activate, action)\nB. (hypothesis, hypotheses)\nC. (to go, went)\nD. (speaking, talking)\n\nAnswer:", | |
| "Question: Which of the following is correct regarding schemas and ontologies?\n\nOptions:\nA. An ontology is created from constructing mappings between schemas\nB. Ontologies can be used for reasoning about different schemas\nC. Ontologies always require a schema\nD. Semi-structured data cannot have a schema\n\nAnswer:", | |
| "Question: Determine which of the following conditional statements evaluate to true (more than one answer can be correct):?\n\nOptions:\nA. If 5-7 > 0, then 3-7<0\nB. If 5-7 < 0, then 3-7<0\nC. If 5-7 > 0, then 3-7>0\nD. If 5-7 < 0, then 3-7>0\n\nAnswer:", | |
| "Question: In which case is the DDH problem believed to be hard?\n\nOptions:\nA. In $\\mathbb{Z}_p^*$.\nB. In $\\mathbb{Z}_n$, where $n = pq$ for two large primes $p$ and $q$.\nC. Given an oracle that returns the least significant bit of $g^{xy}$.\nD. In a large subgroup of prime order of a regular elliptic curve.\n\nAnswer:", | |
| "Question: A model predicts $\\mathbf{\\hat{y}} = [1, 0, 1, 1, 1]$. The ground truths are $\\mathbf{y} = [1, 0, 0, 1, 1]$.\n\nWhat is the accuracy?\n\nOptions:\nA. 0.5\nB. 0.75\nC. 0.8\nD. 0.875\n\nAnswer:", | |
| "Question: Church booleans are a representation of booleans in the lambda calculus. The Church encoding of true and false are functions of two parameters: Church encoding of tru: t => f => t Church encoding of fls: t => f => f What does the following function implement? b => c => b (not c) c?\n\nOptions:\nA. b and c\nB. b or c\nC. b xor c\nD. not c\n\nAnswer:", | |
| "Question: Tick the \\emph{incorrect} assertion when $x\\rightarrow+\\infty$.\n\nOptions:\nA. $x^2+5x+2 = O(x^3)$.\nB. $x^n = O(n^x)$ for any constant $n > 1$.\nC. $x\\log(x) = O(x^2)$.\nD. $1/x = O(1/x^2)$\n\nAnswer:", | |
| "Question: Let $X$, $Y$, and $K$ be respectively the plaintext, ciphertext, and key distributions. $H$ denotes the Shannon entropy. The consequence of perfect secrecy is \\dots?\n\nOptions:\nA. $H(K) \\geq H(X)$\nB. $H(K) \\leq H(X)$\nC. $H(K,X) \\leq H(X)$\nD. $H(Y) \\leq H(X)$\n\nAnswer:", | |
| "Question: Consider a blockcipher $\\mathsf{B}:\\mathcal{K}\\times\\{0,1\\}^n \\rightarrow \\{0,1\\}^n$ with $|\\mathcal{K}|=N$ for which we choose a secret key $K\\in\\mathcal{K}$ uniformly at random. Tick the \\textit{correct} assertion.\n\nOptions:\nA. If $\\mathsf{B}$ is secure against known-plaintext decryption attack, then it is surely secure against ciphertext-only key recovery attack.\nB. If $\\mathsf{B}$ is secure against known-plaintext decryption attack, then it is surely secure against chosen-ciphertext decryption attack.\nC. It is possible to recover $K$ with precomputation: $O(N^{2/3})$, memory: $O(N^{2/3})$, time: $O(1)$.\nD. It is possible to recover $K$ key with precomputation: $O({1})$, memory: $O(N)$, time: $O(1)$.\n\nAnswer:", | |
| "Question: When using bootstrapping in Random Forests, the number of different data items used to construct a single tree is:?\n\nOptions:\nA. smaller than the size of the training data set, with high probability\nB. of order square root of the size of the training set, with high probability\nC. the same as the size of the training data set\nD. subject to the outcome of the sampling process, and can be both smaller or larger than the training set\n\nAnswer:", | |
| "Question: In practice, what is the typical size of an RSA modulus?\n\nOptions:\nA. 64 bits\nB. 256 bits\nC. 1024 bits\nD. 8192 bits\n\nAnswer:", | |
| "Question: Which of the following is correct regarding Crowdsourcing?\n\nOptions:\nA. Random Spammers give always the same answer for every question\nB. It is applicable only for binary classification problems\nC. Honey Pot discovers all the types of spammers but not the sloppy workers\nD. The output of Majority Decision can be equal to the one of Expectation-Maximization\n\nAnswer:", | |
| "Question: Consider an RSA encryption where the public key is published as $(m, e) = (35, 11)$. Which one of the following numbers is a valid decoding exponent?\n\nOptions:\nA. $11$\nB. $7$\nC. $5$\nD. $17$\n\nAnswer:", | |
| "Question: Information extraction:?\n\nOptions:\nA. Necessarily requires training data.\nB. Can be used to populate ontologies.\nC. Is used to identify characteristic entities in a document.\nD. Is always bootstrapped by using ontologies.\n\nAnswer:", | |
| "Question: Select the \\emph{incorrect} statement. Factoring?\n\nOptions:\nA. is a hard problem.\nB. can be solved in polynomial time on a standard computer using Shor's algorithm.\nC. is an easy problem if we have a fast algorithm for computing an order of an element.\nD. is an easy problem if we have a fast algorithm for computing $\\varphi(n)$.\n\nAnswer:", | |
| "Question: Which of the following are parameters involved in the choice made by an order-1 HMM model for PoS tagging knowing that its output isthis/Pron is/V a/Det good/Adj question/Nand that neither \"is\" nor \"question\" can be adjectives, and that \"question\" can also not be a determiner.(Penalty for wrong ticks.)?\n\nOptions:\nA. P(question|N)\nB. P(Adj|V Det)\nC. P(Adj|Det)\nD. P(Det|Adj)\n\nAnswer:", | |
| "Question: Given the following teleporting matrix (\u0395) for nodes A, B and C:\n[0 1\u20442 0]\n[0 0 0]\n[0 1\u20442 1]\nand making no assumptions about the link matrix (R), which of the following is correct:?\n\nOptions:\nA. A random walker can never reach node A\nB. A random walker can never leave node A\nC. A random walker can always leave node C\nD. A random walker can always leave node B\n\nAnswer:", | |
| "Question: Let S(x) be the statement \u201cx has been in a lake\u201d and L(x) be the statement \u201cx lives in Lausanne\u201d and the domain of x consists of all the humans in the world.\n\nOptions:\nA. \nB. S(y) \\wedge \nC. S(y) \\wedge L(y)) \u000bee (x=y) \right] \\Bigr] \\)\nD. \\( \\exists! x (\n\nAnswer:", | |
| "Question: If we pick independent random numbers in $\\{1, 2, \\dots, N\\}$ with uniform distribution, $\\theta \\sqrt{N}$ times, we get at least one number twice with probability\\dots?\n\nOptions:\nA. $e^{\\theta ^2}$\nB. $1-e^{\\theta ^2}$\nC. $e^{-\\theta ^2 /2}$\nD. $1-e^{-\\theta ^2 /2}$\n\nAnswer:", | |
| "Question: Our task is to classify whether an animal is a dog (class 0) or a cat (class 1) based on the following features: \n \begin{itemize}\n\t \\item $x_1$: height\n\t \\item $x_2$: length of whiskers\n\t \\item $x_3$: thickness of fur\n \\end{itemize}\n We perform standard normal scaling on the training features so that they have a mean of zero and standard deviation of 1. We have trained a Logistic Regression model to determine the probability that the animal is a cat, $p(1 | \\mathbf{x,w})$.\n Our classifier learns that cats have a lower height and longer whiskers than dogs, while the thickness of fur is not relevant to the classification outcome. Which of the following is true about the weights~$\\wv$ learned by the classifier?\n ?\n\nOptions:\nA. $w_2 < w_3 < w_1$\nB. $w_1 < w_3 < w_2$\nC. $w_2 < w_1 < w_3$\nD. $w_1 < w_2 < w_3$\n\nAnswer:", | |
| "Question: What is the content of the inode?\n\nOptions:\nA. Filename\nB. Hard links counter\nC. Index structure for data blocks\nD. File mode\n\nAnswer:", | |
| "Question: Let $X$ denote the random variable associated to the plaintexts and $Y$ the random variable associated to the corresponding ciphertexts. If a cryptosystem achieves perfect secrecy, then we have that \\dots?\n\nOptions:\nA. $\\Pr [X=x|Y=y] = \\Pr[Y=y]$.\nB. $\\Pr [X=x|Y=y] = \\Pr[X=x]$.\nC. $\\Pr [X=x|Y=y] = \\Pr[X=x,Y=y]$.\nD. $\\Pr [X=x] = \\Pr[Y=y]$.\n\nAnswer:", | |
| "Question: Select the correct canonical representation(s) of the English verb \"to dance\" conjugated at the third person singular indicative present.\u00a0You will get a penalty for wrong ticks.\n\nOptions:\nA. dances+V+3+S+Ind+Pres\nB. dancing+N+3+S+Ind+Pres\nC. dance+V+3+S+Ind+Pres\nD. dancing+V+3+S+Ind+Pres\n\nAnswer:", | |
| "Question: The edit distance between \u201cpiece\u201d and \u201cpeace\u201d is(Penalty for wrong ticks)?\n\nOptions:\nA. 1, if considering insertion and deletion only\nB. 3\nC. 5\nD. 2, if considering insertion and deletion only\n\nAnswer:", | |
| "Question: The statistical distance between two distributions is \\dots?\n\nOptions:\nA. unrelated to the advantage of a distinguisher.\nB. a lower bound on the advantage of \\emph{all} distinguishers (with a unique sample).\nC. an upper bound on the advantage of \\emph{all} distinguishers (with a unique sample).\nD. an upper bound on the advantage of all distinguishers making statistics on the obtained samples.\n\nAnswer:", | |
| "Question: Which of the following is correct regarding schemas and ontologies?\n\nOptions:\nA. An ontology is created from constructing mappings between schemas\nB. Ontologies can be used for reasoning about different schemas\nC. Ontologies always require a schema\nD. Semi-structured data cannot have a schema\n\nAnswer:", | |
| "Question: Tick the \\textbf{false} assertion. $C(P_0,P_1)\\ldots$?\n\nOptions:\nA. $\\ldots\\displaystyle\\sim-\\log{\\sum_{x}\\sqrt{P_0(x)P_1(x)}}$\nB. $\\ldots=\\displaystyle\\sum_{x}\\left|P_1(x)-P_0(x)\\right|$\nC. $\\ldots\\displaystyle\\sim\\frac{\\mbox{SEI}(P_0,P_1)}{8\\mbox{ln}2}$\nD. is the Chernoff information between $P_0$ ans $P_1$.\n\nAnswer:", | |
| "Question: Given the 2-itemsets {1,2}, {1,5}, {2,5}, {1,4}, {1,3}, when generating the 3-itemsets we\nwill?\n\nOptions:\nA. Generate 5 3-itemsets after the join and 2 3-itemsets after the prune\nB. Generate 6 3-itemsets after the join and 1 3-itemsets after the prune\nC. Generate 4 3-itemsets after the join and 1 3-itemsets after the prune\nD. Generate 4 3-itemsets after the join and 2 3-itemsets after the prune\n\nAnswer:", | |
| "Question: Consider $GF(8)$ defined as $\\mathbb{Z}_2[X]/(P(X))$ with $P(x) = X^3 + X + 1$. Compute $X^2 \\times (X + 1)$ in $\\mathbb{Z}_2[X]/(P(X))$?\n\nOptions:\nA. $X^2+X+1$.\nB. $X^2 + 1$.\nC. $X^2$.\nD. $X+1$.\n\nAnswer:", | |
| "Question: Tick the \\emph{true} assertion among the followings:?\n\nOptions:\nA. Visual cryptography is perfectly secure (at an unreasonable cost).\nB. The Vernam cipher was invented by Kerckoff.\nC. Just like coding theory, cryptography usually faces random noise.\nD. Enigma has never been broken.\n\nAnswer:", | |
| "Question: Consider the Parametric ReLU function defined as\n $$f(x) = \\left\\{\begin{matrix}\n x & \textup{if}\\; x > 0 \\\n ax & \textup{otherwise}\n \\end{matrix}\right.$$\n where $a \\in \\R$ is an arbitrary number.\n Which of the following statements is true regarding the subgradients of $f(x)$ at $x = 0$?\n ?\n\nOptions:\nA. A subgradient exists even though $f(x)$ is not necessarily differentiable at $x=0$.\nB. A subgradient does not exist at $x=0$.\nC. If a subgradient exists, then it is not unique.\nD. None of the mentioned answers.\n\nAnswer:", | |
| "Question: The KNN algorithm needs a notion of distance to assess which points are ``nearest''.\n\t Identify the distance measures that can be used in the KNN algorithm. \n (a) Euclidean Distance : distance associated to the $L_2$ norm $\\|xv\\|_2 := \\sqrt{x_1^2+\\dots+x_D^2}$\n\t\t (b) Manhattan Distance : distance associated to the $L_1$ norm $\\|xv\\|_1 := |x_1|+\\dots+|x_D|$\n\t\t (c) Distance associated to the $L_4$ norm $\\|xv\\|_4 := \big(|x_1|^4+\\dots+|x_D|^4\big)^{1/4}$\n\t ?\n\nOptions:\nA. only a and c\nB. only a and b\nC. a, b and c\nD. only b and c\n\nAnswer:", | |
| "Question: In a second-preimage attack \\dots?\n\nOptions:\nA. you are given a hash function $h$ and have to find two messages $m\\neq m'$ such that $h(m) = h(m')$.\nB. you are given access to $\\text{MAC}_K$ as a blackbox and a tag $\\text{MAC}_K(m)$ for a message $m$ and you have to recover $m$.\nC. you are given a hash function $h$ and a message $m$ and you have to find $m'\\neq m$ such that $h(m) = h(m')$.\nD. you are given a hash function $h$ and $c :=h(h(m))$ for a message $m$ and you have to find $m'$ such that $h(h(m'))=c$.\n\nAnswer:", | |
| "Question: Suppose that an item in a leaf node N exists in every path. Which one is correct?\n\nOptions:\nA. N co-occurs with its prefix in every transaction.\nB. For every node p that is a parent of N in the fp tree, confidence(p->n) = 1 c.\nC. N\u2019s minimum possible support is equal to the number of paths.\nD. The item N exists in every candidate set.\n\nAnswer:", | |
| "Question: Tick all correct answers:?\n\nOptions:\nA. Fuzz testing scales at least to 1'000s of lines of code.\nB. Compiler warnings scale to millions lines of code.\nC. Formal verification scales at least upto 100'000s of lines of code.\nD. Formal verification and concolic execution scale to the same extent.\n\nAnswer:", | |
| "Question: Which of the following attacks makes sense for a MAC:?\n\nOptions:\nA. Chosen message attack.\nB. Chosen ciphertext attack.\nC. Ciphertext only attack.\nD. Decryption attack.\n\nAnswer:", | |
| "Question: Consider the program below. Tick the correct answer. def fun(x: List[Int]) = if x.isEmpty then None else Some(x) val lists = List(List(1, 2, 3), List(), List(4, 5, 6)) for \\t l <- lists \\t v1 <- fun(l) \\t v2 <- fun(v1) yield v2?\n\nOptions:\nA. This program compiles and the last statement has type List[Option[List[Int]]].\nB. This program compiles and the last statement has type List[Option[Int]].\nC. This program compiles and the last statement has type List[Some[Int]].\nD. This program compiles and the last statement has type List[List[Option[Int]]].\n\nAnswer:", | |
| "Question: What adversarial model does not make sense for a message authentication code (MAC)?\n\nOptions:\nA. key recovery.\nB. universal forgery.\nC. existential forgery.\nD. decryption.\n\nAnswer:", | |
| "Question: The Generic Collision Search Algorithm is based on ...\n\nOptions:\nA. the Dictionary Attack.\nB. Birthday Paradox.\nC. the simulation of quantum state on a standard machine.\nD. Murphy's Law.\n\nAnswer:", | |
| "Question: Does the following code snippet contain bugs? If so, which line(s)?\n\nOptions:\nA. Line 6 has a bug / vuln\nB. Line 5 has a bug / vuln\nC. Line 3 has a bug / vuln\nD. Line 4 has a bug / vuln\n\nAnswer:", | |
| "Question: Consider a regression task. You are using your favorite learning algorithm with parameters w and add a regularization term of the form $\\frac{\\lambda}{2}\\|\\mathbf{w}\\|^{2}$. Which of the following statements are correct for a typical scenario?\n\nOptions:\nA. The test error as a function of $\\lambda \\geq 0$ first decreases and then increases.\nB. The training error as a function of $\\lambda \\geq 0$ increases.\nC. The test error as a function of $\\lambda \\geq 0$ increases.\nD. The training error as a function of $\\lambda \\geq 0$ decreases.\n\nAnswer:", | |
| "Question: For $p$ prime and an arbitrary odd $n$, the symbol $\\left( {p \\over n} \\right)$ denotes?\n\nOptions:\nA. a binomial coefficient.\nB. the Legendre symbol.\nC. the Jacobi symbol.\nD. a fraction.\n\nAnswer:", | |
| "Question: Which defense(s) highlight the principle of least privilege in software\n security?\n\nOptions:\nA. DEP bits by disallowing execution on certain memory pages because code is restricted to code pages.\nB. CFI protection on the forward edge because the check limits reachable targets.\nC. Applying updates regularly because software updates always reduce privileges.\nD. A stack canary because it will signal any stack-based attack.\n\nAnswer:", | |
| "Question: Tick the \\textbf{\\emph{incorrect}} assertion.\n\nOptions:\nA. $P\\subseteq NP$.\nB. $NP\\subseteq IP$.\nC. $PSPACE\\subseteq IP$.\nD. $NP\\mbox{-hard} \\subset P$.\n\nAnswer:", | |
| "Question: Tick the \\textbf{true} assertion. Assume an arbitrary $f:\\{0,1\\}^p \\rightarrow \\{0,1\\}^q$, where $p$ and $q$ are integers.\n\nOptions:\nA. $\\mathsf{DP}^f(a,b)=\\displaystyle\\Pr_{X\\in_U\\{0,1\\}^p}[f(X\\oplus a)\\oplus f(X)\\oplus b=1]$, for all $a \\in \\{0,1\\}^p$, $b \\in \\{0,1\\}^q$.\nB. $\\Pr[f(x\\oplus a)\\oplus f(x)\\oplus b=0]=E(\\mathsf{DP}^f(a,b))$, for all $a, x \\in \\{0,1\\}^p$, $b \\in \\{0,1\\}^q$.\nC. $2^p\\mathsf{DP}^f(a,b)$ is odd, for all $a \\in \\{0,1\\}^p, b \\in \\{0,1\\}^q$.\nD. $\\displaystyle\\sum_{b\\in\\{0,1\\}^q} \\mathsf{DP}^f(a,b)=1$, for all $a \\in \\{0,1\\}^p$.\n\nAnswer:", | |
| "Question: How many generators are there in $\\mathbb{Z}_n$?\n\nOptions:\nA. $1$\nB. $n-1$\nC. $n$\nD. $\\varphi (n)$\n\nAnswer:", | |
| "Question: You share an apartment with friends.\nKitchen, living room, balcony, and bath room are shared resources among all parties.\nWhich policy/policies violate(s) the principle of least privilege?\n\nOptions:\nA. Different bedrooms do not have a different key.\nB. There is no lock on the fridge.\nC. To access the kitchen you have to go through the living room.\nD. Nobody has access to the neighbor's basement.\n\nAnswer:", | |
| "Question: What is the order of 11 in $\\mathbb{Z}_{37}$?\n\nOptions:\nA. 1\nB. 27\nC. 36\nD. 37\n\nAnswer:", | |
| "Question: Let $0\\leq\\alpha\\leq1$ be an unknown constant. Let $X$ be a random variable taking values in $\\mathcal{X}=\\{0,1,2\\}$ with probability $p_X(0) = p_X(1) = \\alpha$ and $p_X(2) = 1-2\\alpha$. Let $Y$ be a random variable defined as follows \begin{equation*}Y = \begin{cases}1, &\text{if }X=2\\0, &\text{if } X\neq 2\\end{cases}.\\end{equation*} You also know that $H(X|Y) = \frac{1}{2}$. Choose the correct value of $\\alpha$.\n\nOptions:\nA. $\frac{1}{4}$\nB. $\frac{1}{2}$\nC. $1$\nD. $\frac{1}{8}$\n\nAnswer:", | |
| "Question: In classification, which of the following is true regarding class imbalance?\n\nOptions:\nA. Classes should have the same distribution in the validation set and in the full dataset.\nB. Oversampling the larger class can reduce the impact of the skewed distribution.\nC. Oversampling rare classes in the testing set can reduce the impact of skewed distribution.\nD. The leave-one-out methodology produces the same class distribution in the training and the testing set.\n\nAnswer:", | |
| "Question: For binary classification, which of the following methods can achieve perfect training accuracy on \\textbf{all} linearly separable datasets?\n\nOptions:\nA. Decision tree\nB. Hard-margin SVM\nC. 15-nearest neighbors\nD. None of the suggested\n\nAnswer:", | |
| "Question: Tick the \\emph{false} assertion. Using double encryption leads to\\dots?\n\nOptions:\nA. doubling the complexity of the Encryption algorithm.\nB. doubling the complexity of the Decryption algorithm.\nC. doubling the complexity of all the attacks against the cipher.\nD. doubling the length of the key.\n\nAnswer:", | |
| "Question: One of the following ciphers is a \\emph{block} cipher. Which one?\n\nOptions:\nA. AES\nB. RC4\nC. A5/1\nD. MD5\n\nAnswer:", | |
| "Question: Modularity clustering will end up always with a single community at the top level?\n\nOptions:\nA. true\nB. Only for dense graphs\nC. Only for connected graphs\nD. never\n\nAnswer:", | |
| "Question: Let $\\mathcal{C}$ be a $(n,k)$ Reed-Solomon code on $\\mathbb{F}_q$. Let $\\mathcal{C}'$ be the $(2n,k)$ code such that each codeword of $\\mathcal{C}'$ is a codeword of $\\mathcal{C}$ repeated twice, i.e., if $(x_1,\\dots,x_n) \\in\\mathcal{C}$, then $(x_1,\\dots,x_n,x_1,\\dots,x_n)\\in\\mathcal{C'}$. What is the minimum distance of $\\mathcal{C}'$?\n\nOptions:\nA. $2n-2k+2$\nB. $2n-k+1$\nC. $2n-2k+1$\nD. $2n-k+2$\n\nAnswer:", | |
| "Question: Consider the loss function $L: \\R^d \to \\R$, $L(\\wv) = \frac{\beta}{2}\\|\\wv\\|^2$, where $\beta > 0$ is a constant. We run gradient descent on $L$ with a stepsize $\\gamma > 0$ starting from some $\\wv_0 \neq 0$. Which of the statements below is true? ?\n\nOptions:\nA. Gradient descent converges to the global minimum for any stepsize $\\gamma > 0$.\nB. Gradient descent with stepsize $\\gamma = \frac{2}{\beta}$ produces iterates that diverge to infinity ($\\|\\wv_t\\| \to \\infty$ as $t\to \\infty$).\nC. Gradient descent converges in two steps for $\\gamma = \frac{1}{\beta}$ (i.e., $\\wv_2$ is the \textbf{first} iterate attaining the global minimum of $L$).\nD. Gradient descent converges to the global minimum for any stepsize in the interval $\\gamma \\in \big( 0, \frac{2}{\beta}\big)$.\n\nAnswer:", | |
| "Question: Suppose that an item in a leaf node N exists in every path. Which one is correct? ?\n\nOptions:\nA. N co-occurs with its prefix in every transaction.\nB. For every node P that is a parent of N in the fp tree, confidence(P->N) = 1\nC. N\u2019s minimum possible support is equal to the number of paths.\nD. The item N exists in every candidate set.\n\nAnswer:", | |
| "Question: Tick the \\emph{correct} assertion. In RSA with public key $(e,N)$ and private key $(d,N)$ \\ldots?\n\nOptions:\nA. the key generation requires that $\\gcd(e,N) = 1$.\nB. decryption does not work for some $y \\in \\mathbb{Z}_N$ for which $\\gcd(y,N) \\not= 1$.\nC. if we know $\\phi(N)$, the ciphertext $y$ and the public key, we can decrypt $y$.\nD. in order to decrypt we need to have $e^d \\equiv 1 \\pmod{N}$.\n\nAnswer:", | |
| "Question: Tick the \\textbf{false} assertion. The SEI of the distribution $P$ of support $G$ \\ldots?\n\nOptions:\nA. is equal to \\# $G\\cdot\\displaystyle\\sum_{x\\in G}\\left(P(x)-\\frac{1}{\\sharp G}\\right)^2$\nB. is the advantage of the best distinguisher between $P$ and the uniform distribution.\nC. denotes the Squared Euclidean Imbalance.\nD. is positive.\n\nAnswer:", | |
| "Question: Consider the following context-free grammar \\(G\\) (where \\(\\text{S}\\) is the top-level symbol):\n\nOptions:\nA. 0\nB. 2\nC. 4\nD. 1\n\nAnswer:", | |
| "Question: Tick the \\emph{incorrect} assertion. In \\emph{all} finite Abelian groups $(G,+)$ \\ldots?\n\nOptions:\nA. the order of any element divides the order of the group.\nB. all the elements are generators.\nC. the operation $+$ is commutative.\nD. the neutral element has order $1$.\n\nAnswer:", | |
| "Question: Fundamentally, why clustering is considered an unsupervised machine learning technique?\n\nOptions:\nA. Number of clusters are not known.\nB. The class labels are not known.\nC. The features are not known.\nD. The clusters can be different with different initial parameters.\n\nAnswer:", | |
| "Question: The edit distance between \u201cpiece\u201d and \u201cpeace\u201d is(Penalty for wrong ticks)?\n\nOptions:\nA. 1, if considering insertion, deletion and substitution\nB. 2, if considering insertion, deletion and substitution\nC. 3\nD. 3, if considering insertion and deletion only\n\nAnswer:", | |
| "Question: In which of the following groups is the decisional Diffie-Hellman problem (DDH) believed to be hard?\n\nOptions:\nA. In $\\mathbb{Z}_p$, with a large prime $p$.\nB. In large subgroup of smooth order of a ``regular'' elliptic curve.\nC. In a large subgroup of prime order of $\\mathbb{Z}_p^*$, such that $p$ is a large prime.\nD. In $\\mathbb{Z}_p^*$, with a large prime $p$.\n\nAnswer:", | |
| "Question: Select what is true about the Baum-Welch algorithm.A penalty will be applied for any incorrect answers.\n\nOptions:\nA. It uses expectation maximization to estimate parameters of the Hidden Markov Model.\nB. The algorithm always converges towards a global maximum of the parameters estimated.\nC. The resulting parameters from the algorithm do not depend on the values the parameters are initialized with.\nD. If we have few word sequences annotated with PoS tags, we can use these data to set, in a supervised manner, the initial parameters of the Baum-Welch algorithm.\n\nE. You cannot randomly initialize the parameters of the algorithm.\n\nAnswer:", | |
| "Question: To which expression is the following for-loop translated? for x <- xs if x > 5; y <- ys yield x + y?\n\nOptions:\nA. xs.flatMap(x => ys.map(y => x + y)).withFilter(x => x > 5)\nB. xs.withFilter(x => x > 5).map(x => ys.flatMap(y => x + y))\nC. xs.withFilter(x => x > 5).flatMap(x => ys.map(y => x + y))\nD. xs.map(x => ys.flatMap(y => x + y)).withFilter(x => x > 5)\n\nAnswer:", | |
| "Question: Your friend Charlie was training a neural net, and observed something very curious. During training, the same network, with the same exact weights, doesn't always give the same prediction for a given example. What can cause this behavior?\n\nOptions:\nA. Batch normalization\nB. Residual layers (skip-connections)\nC. Dropout\nD. Convolutional layers\n\nAnswer:", | |
| "Question: In a MAC forgery, the adversary tries to\\dots?\n\nOptions:\nA. recover the message $m$ given the tag \\textsf{MAC}_K(m).\nB. decrypt a given message.\nC. forge the secret key.\nD. compute the MAC of a message whose MAC was never computed before.\n\nAnswer:", | |
| "Question: Which one is \\emph{not} a block cipher?\n\nOptions:\nA. DES\nB. ECB\nC. SAFER\nD. IDEA\n\nAnswer:", | |
| "Question: Tick the \\emph{incorrect} assertion.\n\nOptions:\nA. Commitment schemes never use randomness.\nB. A commitment scheme can be based on the hardness of the discrete logarithm problem.\nC. A commitment scheme should be hiding and binding.\nD. Perdersen Commitment uses two large primes.\n\nAnswer:", | |
| "Question: Which of the following statements about libFuzzer is/are correct?\n\nOptions:\nA. In libFuzzer\u2019s default mode (not fork-mode), the tested APIs\nB. must not contain \\texttt{exit()}.\nC. stubs.\nD. Unit tests may serve as foundation to create libFuzzer fuzzing\n\nAnswer:", | |
| "Question: Tick the \\textbf{minimal} requirements on a channel for performing a Diffie-Hellman key exchange securely. The channel must\\dots?\n\nOptions:\nA. be authenticated, confidential and protect the integrity.\nB. protect the integrity.\nC. be confidential.\nD. be authenticated and protect the integrity.\n\nAnswer:", | |
| "Question: Let $n=pq$ be a RSA modulus and let $(e,d)$ be a RSA public/private key. Tick the \\emph{correct} assertion.\n\nOptions:\nA. Finding a multiple of $\\lambda(n)$ is equivalent to decrypt a ciphertext.\nB. $ed$ is a multiple of $\\phi(n)$.\nC. The two roots of the equation $X^2 - (n-\\phi(n)+1)X+n$ in $\\mathbb{Z}$ are $p$ and $q$.\nD. $e$ is the inverse of $d$ mod $n$.\n\nAnswer:", | |
| "Question: Tick the \\textbf{true} assertion. The advantage of a distinguisher of two distributions $P_0$ and $P_1$?\n\nOptions:\nA. is always the Euclidean distance between $P_0$ and $P_1$.\nB. is $\\mathsf{Adv}_{\\mathcal{A}} (P_0 , P_1 ) = \\Pr[P = P_1|A \\rightarrow 1]-\\Pr[P = P_0| A \\rightarrow 1]$.\nC. is $\\mathsf{Adv}_{\\mathcal{A}} (P_0 , P_1 ) = \\Pr[A \\rightarrow 0|P = P_1 ]-\\Pr[A \\rightarrow 1|P = P_0]$.\nD. can touch the statistical distance $\\frac{1}{2}\\Sigma_{x}|P_0(x) - P_1(x)|$ between $P_0$ and $P_1$, when he makes only one query.\n\nAnswer:", | |
| "Question: Consider the composite function $f(x)=g(h(x))$, where all functions are $\\mathbb{R}$ to $\\mathbb{R}$. Which of the following is the weakest condition that guarantees that $f(x)$ is convex?\n\nOptions:\nA. $g(x)$ is convex and $g(x)$ is increasing\nB. $g(x)$ and $h(x)$ are convex and $h(x)$ is increasing\nC. $g(x)$ and $h(x)$ are convex and $g(x)$ and $h(x)$ are increasing\nD. $g(x)$ and $h(x)$ are convex and $g(x)$ is increasing\n\nAnswer:", | |
| "Question: Which one of the following notions is not in the fundamental trilogy of cryptography?\n\nOptions:\nA. authentication\nB. confidentiality\nC. integrity\nD. privacy\n\nAnswer:", | |
| "Question: For the number of times the apriori algorithm and the FPgrowth algorithm for association rule mining are scanning the transaction database the following is true?\n\nOptions:\nA. fpgrowth has always strictly fewer scans than apriori\nB. fpgrowth and apriori can have the same number of scans\nC. apriori cannot have fewer scans than fpgrowth\nD. all three above statements are false\n\nAnswer:", | |
| "Question: Tick the \\emph{false} assertion. Given a ring $R$, $R^\\star$ is\\ldots?\n\nOptions:\nA. a group.\nB. the set of invertible elements in $R$.\nC. $R-\\{0\\}$.\nD. the set of units.\n\nAnswer:", | |
| "Question: Consider a group $G$ with prime order. We can deduce that?\n\nOptions:\nA. all elements are generators.\nB. all elements are generators except the neutral element.\nC. half of the elements are generators.\nD. $G$ is a field.\n\nAnswer:", | |
| "Question: When computing PageRank iteratively, the computation ends when:?\n\nOptions:\nA. The norm of the difference of rank vectors of two subsequent iterations falls below a predefined threshold\nB. The difference among the eigenvalues of two subsequent iterations falls below a predefined threshold\nC. All nodes of the graph have been visited at least once\nD. The probability of visiting an unseen node falls below a predefined threshold\n\nAnswer:", | |
| "Question: What is TRUE regarding the Multi-head Self-attention mechanism?\n\nOptions:\nA. Its computation cannot be parallelized compared to LSTMs and other sequential models.\nB. It gives the Transformer the ability to learn different aspects of the meanings of each word.\nC. LSTMs have larger memory than models with self-attention.\nD. Its complexity is linear to the input size.\n\nAnswer:", | |
| "Question: Recall can be defined as:?\n\nOptions:\nA. P(relevant documents | retrieved documents)\nB. P(retrieved documents relevant documents)\nC. P(retrieved documents number of documents)\nD. P(relevant documents number of documents)\n\nAnswer:", | |
| "Question: Let us define the relation R = {(1, 1)} on the set A = {1, 2, 3}. Which of the following properties does R satisfy ? (multiple answers)?\n\nOptions:\nA. Reflexive\nB. Symmetric\nC. Transitive\nD. Anti-symmetric\n\nAnswer:", | |
| "Question: Select the most accurate answer. For an RSA key to have a good security level in 2015, the modulus should be of length\\dots?\n\nOptions:\nA. around 80 bits.\nB. around 1300 bits.\nC. around 160 bits.\nD. around 512 bits.\n\nAnswer:", | |
| "Question: Let $E$ be an elliptic curve. Solving which of the following problems would help you to break Elliptic Curve Diffie-Hellman (ECDH) over $E$?\n\nOptions:\nA. Let $P,R \\in E$. Given $P$ and $P+R$, compute $R$.\nB. Let $P,Q \\in E$. Given $P$ and $Q$, compute the product between $P$ and $Q$, i.e., $P \\times Q$.\nC. Let $P \\in E$ and $\\ell \\in \\mathbb{N}$. Given $P$ and $\\ell P$, compute $\\ell$.\nD. Find a point which is not on the curve $E$.\n\nAnswer:", | |
| "Question: Find all solutions of $24x + [9]_{45} = [13]_{45}$ in the range $[0,44]$. How many different solutions are there?\n\nOptions:\nA. $0$\nB. $1$\nC. $2$\nD. $3$\n\nAnswer:", | |
| "Question: Let us assume that a kernel $K: \\mathcal{X} \\times \\mathcal{X} \\rightarrow \\mathbb{R}$ is said to be valid if there exists $k \\in \\mathbb{N}$ and $\\Phi: \\mathcal{X} \\rightarrow \\mathbb{R}^{k}$ such that for all $\\left(x, x^{\\prime}\\right) \\in \\mathcal{X} \\times \\mathcal{X}, K\\left(x, x^{\\prime}\\right)=\\Phi(x)^{\\top} \\Phi\\left(x^{\\prime}\\right)$ Which one of the following kernels is not valid ?\n\nOptions:\nA. $\\mathcal{X}=\\mathbb{Q}, K\\left(x, x^{\\prime}\\right)=2^{x+x^{\\prime}}$\nB. $\\mathcal{X}=\\mathbb{N}, K\\left(x, x^{\\prime}\\right)=2$\nC. $\\mathcal{X}=\\mathbb{R}^{d}, K\\left(x, x^{\\prime}\\right)=\\left(x^{\\top} x^{\\prime}\\right)^{2}$\nD. $\\mathcal{X}=\\mathbb{R}, K\\left(x, x^{\\prime}\\right)=\\cos \\left(x-x^{\\prime}\\right)$\n\nAnswer:", | |
| "Question: We apply a Gaussian Mixture Model made of $K$ isotropic Gaussians (invariant to rotation around its center) to $N$ vectors of dimension $D$. What is the number of \\emph{free} parameters of this model?\n ?\n\nOptions:\nA. $KD + 2K - 1 + N - 1$\nB. $KD + K - 1$\nC. $KD + 2K - 1 - 1$\nD. $KD + KD^2 - 1$\n\nAnswer:", | |
| "Question: Which of the following provides forward secrecy.\n\nOptions:\nA. Transferring keys in plain.\nB. Ephemeral Diffie-Hellman.\nC. Semi-static Diffie-Hellman.\nD. Static Diffie-Hellman.\n\nAnswer:", | |
| "Question: Which of the following statement(s) is/are true? To achieve memory safety for C, ...\n\nOptions:\nA. a strict specification of program functionality is required\nB. pointer arithmetic must be checked/restricted\nC. all control-flow must be validated against a model\nD. memory deallocation must be replaced by garbage collection\n\nAnswer:", | |
| "Question: What is the order of $2^{124}$ in $(\\mathbb{Z}_{2^{128}},+)$?\n\nOptions:\nA. 8.\nB. $\\varphi(2^{128})$.\nC. 124.\nD. 16.\n\nAnswer:", | |
| "Question: A query \\(q\\) has been submitted to two distinct Information Retrieval engines operating on the same document collection containing 1'000 documents, with 50 documents being truly relevant for \\(q\\).The following result lists have been produced by the two IR engines, \\(S_1\\) and \\(S_2\\) respectively:\n\n\\(S_1\\text{:}\\)\n\\(d_1\\)\n\\(d_2\\text{ (*)}\\)\n\\(d_3\\text{ (*)}\\)\n\\(d_4\\)\n\\(d_5\\text{ (*)}\\)\n\n\\(S_2\\text{:}\\)\n\\(d^\\prime_1\\text{ (*)}\\)\n\\(d^\\prime_2\\text{ (*)}\\)\n\\(d^\\prime_3\\)\n\\(d^\\prime_4\\)\n\\(d^\\prime_5\\)\n\nIn these result lists, the stars \\(\\text{(*)}\\) identify the truly relevant documents.\nBy convention, we consider that any non retrieved document has been retrieved at rank 6.If Average Precision is used as evaluation metric, which of the two IR engines is performing better for the query \\(q\\)?\n\nOptions:\nA. \\(S_1\\)\nB. \\(S_2\\)\nC. Both engines perform equally.\nD. This evaluation metric cannot be computed.\n\nAnswer:", | |
| "Question: $\\mathbb{Z}_{37}^*$ denotes ...\n\nOptions:\nA. a field.\nB. an additive group.\nC. a multiplicative group.\nD. a ring.\n\nAnswer:", | |
| "Question: When constructing a word embedding, what is TRUE regarding negative samples?\n\nOptions:\nA. They are oversampled if less frequent\nB. Their frequency is decreased down to its logarithm\nC. They are words that do not appear as context words\nD. They are selected among words that are not stop-words\n\nAnswer:", | |
| "Question: Consider a binary classification problem with classifier $f(\\mathbf{x})$ given by $$ f(\\mathbf{x})= \\begin{cases}1, & g(\\mathbf{x}) \\geq 0 \\\\ -1, & g(\\mathbf{x})<0\\end{cases} $$ and $\\mathbf{x} \\in \\mathbb{R}^{6}$. Consider a specific pair $(\\mathbf{x}, y=1)$ and assume that $g(\\mathbf{x})=8$. In particular this means that this point is classified correctly by $f$. Assume further that we have computed the gradient of $g$ at $\\mathbf{x}$ to be $\\nabla_{\\mathbf{x}} g(\\mathbf{x})=(+1,-2,+3,-4,+5,-6)$. You are allowed to make one step in order to (hopefully) find an adversarial example. In the following four questions, assume $\\epsilon=1$. What is the value of $g(\\mathbf{x}+\\delta)$ for this $\\ell_{\\infty}$-optimal choice assuming that $g$ is (locally) linear?\n\nOptions:\nA. $+7$\nB. $4$\nC. $0$\nD. $+13$\n\nAnswer:", | |
| "Question: Tick the \\emph{incorrect} assertion. The security of the DH protocol requires \\dots?\n\nOptions:\nA. that the DL problem is hard.\nB. that the factoring problem is hard.\nC. that we use groups of prime order.\nD. a KDF function.\n\nAnswer:", | |
| "Question: Which of the following is true for Recommender Systems (RS)?\n\nOptions:\nA. Matrix Factorization can predict a score for any user-item combination in the dataset.\nB. The complexity of the Content-based RS depends on the number of users\nC. Item-based RS need not only the ratings but also the item features\nD. Matrix Factorization is typically robust to the cold-start problem.\n\nAnswer:", | |
| "Question: Consider the following joint distribution on $X$ and $Y$, where both random variables take on the values $\\{0,1\\}: p(X=$ $0, Y=0)=0.1, p(X=0, Y=1)=0.2, p(X=1, Y=0)=0.3, p(X=1, Y=1)=0.4$. You receive $X=1$. What is the largest probability of being correct you can achieve when predicting $Y$ in this case?\n\nOptions:\nA. $\\frac{3}{4}$\nB. $\\frac{4}{7}$\nC. $\\frac{1}{7}$\nD. $\\frac{1}{3}$\n\nAnswer:", | |
| "Question: What is the minimum distance of a linear block code over $\\mathbb{F}_{7}$ that has $$H = \\left(\begin{array}{cccccc}1&0&0&0&3&6\\0&1&0&0&6&6\\0&0&1&0&6&3\\end{array}\right)$$ as the parity check matrix? Check the correct answer.\n\nOptions:\nA. 0\nB. 3\nC. 2\nD. 1\n\nAnswer:", | |
| "Question: Tick the \\emph{correct} assertion regarding the Diffie-Hellman key exchange over a group $G$.\n\nOptions:\nA. It is never secure against a passive adversary.\nB. It is not secure if you apply a key derivation function to the output of the exchange.\nC. It is not secure when discrete logarithms in $G$ can be computed efficiently.\nD. It is always secure if the order of $G$ is a prime number.\n\nAnswer:", | |
| "Question: If we run the Apriori algorithm on the following transaction database with minimal support count of 2, which of the itemsets will have a support count of 3?Transaction IDItemsT1{1,3,4}T2{2,3,5}T3{1,2,3,5}T4{2,5}T5{1,2,3,5}?\n\nOptions:\nA. {1,5}\nB. {2,3}\nC. {2,5}\nD. {1,2}\n\nAnswer:", | |
| "Question: Which one is a stream cipher?\n\nOptions:\nA. RSA\nB. MD5\nC. AES\nD. RC4\n\nAnswer:", | |
| "Question: We saw in class that we can quickly decrease the spatial size of the representation using pooling layers. Is there another way to do this without pooling?\n\nOptions:\nA. Yes, by increasing the amount of padding.\nB. Yes, by increasing the stride.\nC. Yes, by increasing the number of filters.\nD. No, pooling is necessary.\n\nAnswer:", | |
| "Question: A simple substitution cipher can be broken \\dots?\n\nOptions:\nA. by analysing the probability occurence of the language.\nB. only by using a quantum computer.\nC. by using the ENIGMA machine.\nD. by using public-key cryptogaphy.\n\nAnswer:", | |
| "Question: We are given a data set $S=\\left\\{\\left(\\boldsymbol{x}_{n}, y_{n}\\right)\\right\\}$ for a binary classification task where $\\boldsymbol{x}_{n}$ in $\\mathbb{R}^{D}$. We want to use a nearestneighbor classifier. In which of the following situations do we have a reasonable chance of success with this approach? [Ignore the issue of complexity.]?\n\nOptions:\nA. $n \\rightarrow \\infty, D$ is fixed\nB. $ n \\rightarrow \\infty, D \\ll \\ln (n)$\nC. $ n=D^2, D \\rightarrow \\infty$\nD. $ n$ is fixed, $D \\rightarrow \\infty$\n\nAnswer:", | |
| "Question: Why is XML a document model?\n\nOptions:\nA. It supports application-specific markup\nB. It supports domain-specific schemas\nC. It has a serialized representation\nD. It uses HTML tags\n\nAnswer:", | |
| "Question: A retrieval model attempts to capture?\n\nOptions:\nA. the interface by which a user is accessing information\nB. the importance a user gives to a piece of information for a query\nC. the formal correctness of a query formulation by user\nD. the structure by which a document is organised\n\nAnswer:", | |
| "Question: Decision trees...\n\nOptions:\nA. ... can be used for both classification and regression.\nB. ... can be easily explained.\nC. ... need water and sunlight to grow.\nD. ... have several different roots.\n\nAnswer:", | |
| "Question: A colleague challenges you to create a $(n-1,k,d_{min})$ code $\\mathcal C'$ from a $(n,k,d_{min})$ code $\\mathcal C$ as follows: given a generator matrix $G$ that generates $\\mathcal C$, drop one column from $G$. Then, generate the new code with this truncated $k \times (n-1)$ generator matrix. The catch is that your colleague only gives you a set $\\mathcal S=\\{\\vec s_1,\\vec s_2, \\vec s_3\\}$ of $3$ columns of $G$ that you are allowed to drop, where $\\vec s_1$ is the all-zeros vector, $\\vec s_2$ is the all-ones vector, and $\\vec s_3$ is a canonical basis vector. From the length of the columns $s_i$ you can infer $k$. You do not know $n$, neither do you know anything about the $n-3$ columns of $G$ that are not in $\\mathcal S$. However, your colleague tells you that $G$ is in systematic form, i.e., $G=[I ~~ P]$ for some unknown $P$, and that all of the elements in $\\mathcal S$ are columns of $P$. Which of the following options in $\\mathcal S$ would you choose as the column of $G$ to drop?\n\nOptions:\nA. $\\vec s_1$ (the all-zeros vector).\nB. $\\vec s_2$ (the all-ones vector)\nC. $\\vec s_3$ (one of the canonical basis vectors).\nD. It is impossible to guarantee that dropping a column from $\\mathcal S$ will not decrease the minimum distance.\n\nAnswer:", | |
| "Question: I want to send a value to Bob without him knowing which value I sent and such that I cannot change my mind later when I reveal it in clear. I should use \\dots?\n\nOptions:\nA. a stream cipher.\nB. a PRNG.\nC. a commitment scheme.\nD. a digital signature.\n\nAnswer:", | |
| "Question: Tick the \\emph{incorrect} assertion. A distinguisher \\dots?\n\nOptions:\nA. can be used to break a PRNG.\nB. is an algorithm calling an oracle.\nC. can prove the security of a block cipher.\nD. can show a weakness in a design.\n\nAnswer:", | |
| "Question: Which of the following are parameters involved in the choice made by an order-1 HMM model for PoS tagging knowing that its output isthis/Pron is/V a/Det good/Adj question/Nand that neither \"is\" nor \"question\" can be adjectives, and that \"question\" can also not be a determiner.(Penalty for wrong ticks.)?\n\nOptions:\nA. P(this is)\nB. P(this V)\nC. P(question|N)\nD. P(this)\n\nAnswer:", | |
| "Question: Dude said \u201cI like bowling\u201d. With how many statements can we express this sentence using \u200b RDF Reification?\n\nOptions:\nA. We cannot\nB. 1\nC. 3\nD. 5\n\nAnswer:", | |
| "Question: What should be the type of value ? // list: List[String] list.foldLeft(value)((x, y) => x + y.length)?\n\nOptions:\nA. List[Int]\nB. Int\nC. List[String]\nD. List[Char]\n\nAnswer:", | |
| "Question: Tick the \\textbf{false} assertion. Vernam is perfectly secure when \\ldots?\n\nOptions:\nA. \\ldots the key is at least as long as the message\nB. \\ldots a key is never used twice\nC. \\ldots the key is a random prime number\nD. \\ldots the key source is truly random\n\nAnswer:", | |
| "Question: Given the following lemmas, holding for all types A, x: A, b1: Bool, b2: Bool, p: A => Bool, xs: List[A] and ys: List[A]: (ForallNil) nil.forall(p) === True (ForallCons) (x :: xs).forall(p) === p(x) && xs.forall(p) (ExistsNil) nil.exists(p) === False (ExistsCons) (x :: xs).exists(p) === p(x) || xs.exists(p) (NegFalse) !False === True (NegOr) !(b1 || b2) === !b1 && !b2 (NegAnd) !(b1 && b2) === !b1 || !b2 (NegInvolutive) !!b1 === b1 Let us prove the following lemma for all l: List[A] and all p: A => Bool: (ListNegExists) !l.exists(x => !p(x)) === l.forall(p) We prove it by induction on l. Base case: l is Nil. Therefore, we need to prove: !Nil.exists(x => !p(x)) === Nil.forall(p) Starting from the left hand-side (!Nil.exists(x => !p(x))), what exact sequence of lemmas should we apply to get the right hand-side (Nil.forall(p))?\n\nOptions:\nA. NegInvolutive, ForallNil, ExistsNil\nB. NegFalse, ForallNil, ExistsNil\nC. NegFalse, ExistsNil, ForallNil\nD. ForallNil, NegFalse, ExistsNil\n\nAnswer:", | |
| "Question: Consider an RSA encryption where the public key is published as $(m, e) = (55, 17)$. Which one of the following numbers is a valid decoding exponent?\n\nOptions:\nA. $53$\nB. $23$\nC. $43$\nD. $83$\n\nAnswer:", | |
| "Question: A 128-bit key ...\n\nOptions:\nA. has 128 decimal digits.\nB. is too long for any practical application.\nC. provides reasonable security for at least four decades.\nD. adresses $n^2$ problem for $n=2^{64}$.\n\nAnswer:", | |
| "Question: Which of the following statements is wrong regarding RDF?\n\nOptions:\nA. An RDF statement would be expressed in SQL as a tuple in a table\nB. Blank nodes in RDF graphs correspond to the special value NULL in SQL\nC. The object value of a type statement corresponds to a table name in SQL\nD. RDF graphs can be encoded as SQL databases\n\nAnswer:", | |
| "Question: The Vernam cipher \\ldots?\n\nOptions:\nA. is always secure.\nB. is secure only if we always use the same key.\nC. is secure only if we always use a different key.\nD. is always insecure.\n\nAnswer:", | |
| "Question: Tick the \\textbf{true} statement regarding $\\mathsf{GF}(2^k)$.\n\nOptions:\nA. Addition is performed as bitwise XOR.\nB. Multiplication is performed modulo $2^k$.\nC. It has order $\\varphi(2^k)$.\nD. Its characteristic is $k$.\n\nAnswer:", | |
| "Question: The edit distance between \u201cpiece\u201d and \u201cpeace\u201d is(Penalty for wrong ticks)?\n\nOptions:\nA. 3\nB. 3, if considering insertion, deletion and substitution\nC. 1, if considering insertion, deletion, transposition and substitution\nD. 2, if considering insertion, deletion, transposition and substitution\n\nAnswer:", | |
| "Question: What is WRONG\u00a0regarding the Transformer model?\n\nOptions:\nA. It uses a self-attention mechanism to compute representations of the input and output.\nB. Its computation cannot be parallelized compared to LSTMs and other sequential models.\nC. Its complexity is quadratic to the input size.\nD. It captures the semantic context of the input.\n\nAnswer:", | |
| "Question: The worst case complexity of an exaustive search (with memory) against DES is\\dots?\n\nOptions:\nA. $1$\nB. $\\frac{2^{64}}{2}$\nC. $2^{56}$\nD. $2^{64}$\n\nAnswer:", | |
| "Question: Let $n$ be an integer. What is the cardinality of $\\mathbf{Z}^*_n$?\n\nOptions:\nA. $n$\nB. $n-1$\nC. $\\varphi(n)$\nD. $\\varphi(n-1)$\n\nAnswer:", | |
| "Question: The encryption in GSM is done by ...\n\nOptions:\nA. A3.\nB. A8.\nC. E0.\nD. A5.\n\nAnswer:", | |
| "Question: The KNN algorithm needs a notion of distance to assess which points are ``nearest''.\n\t Identify the distance measures that can be used in the KNN algorithm. \n (a) Euclidean Distance : distance associated to the $L_2$ norm $\\|xv\\|_2 := \\sqrt{x_1^2+\\dots+x_D^2}$\n\t\t (b) Manhattan Distance : distance associated to the $L_1$ norm $\\|xv\\|_1 := |x_1|+\\dots+|x_D|$\n\t\t (c) Distance associated to the $L_4$ norm $\\|xv\\|_4 := \big(|x_1|^4+\\dots+|x_D|^4\big)^{1/4}$\n\t ?\n\nOptions:\nA. only b\nB. a, b and c\nC. only c\nD. only a\n\nAnswer:", | |
| "Question: Select \\emph{incorrect} statement.\n\nOptions:\nA. The high entropy of password is ensured by hashing the password before storing.\nB. A random salt provides some level of protection against dictionary attack.\nC. A password of length $8$ characters should contain non-alphanumerical characters in order to be non-recoverable by brute force.\nD. A special purpose device can recover $56$ bit passwords in a matter of minutes.\n\nAnswer:", | |
| "Question: Consider a twelve-round Feistel scheme $\\Psi(F^{k_1},F^{k_2},\\dots,F^{k_{12}})$. Tick the \\emph{minimal necessary} condition for $\\Psi$ to be a permutation:?\n\nOptions:\nA. $F^{k_i}$ are functions\nB. $F^{k_i}$ are injective functions\nC. $F^{k_i}$ are surjective functions\nD. $F^{k_i}$ are bijective functions\n\nAnswer:", | |
| "Question: Tick the \\textbf{true} assertion. In RSA \\ldots?\n\nOptions:\nA. \\ldots decryption is known to be equivalent to factoring.\nB. \\ldots key recovery is provably not equivalent to factoring).\nC. \\ldots decryption is probabilistic.\nD. \\ldots public key transmission needs authenticated and integer channel.\n\nAnswer:", | |
| "Question: The Tonelli algorithm is?\n\nOptions:\nA. a factoring algorithm.\nB. a primality testing algorithm.\nC. an algorithm for finding square roots.\nD. an algorithm for testing quadratic residuosity.\n\nAnswer:", | |
| "Question: Which of the following is/are true about fuzzing with sanitizers?\n\nOptions:\nA. The set of sanitizers used during a fuzzing campaign must be\nB. carefully chosen (tradeoff between bug visibility/execution\nC. speed).\nD. Some fuzzers use fork servers to reduce sanitizer overhead.\n\nAnswer:", | |
| "Question: The type statement in RDF would be expressed in the relational data model by a table?\n\nOptions:\nA. with one attribute\nB. with two attributes\nC. with three attributes\nD. cannot be expressed in the relational data model\n\nAnswer:", | |
| "Question: Tick the \\textbf{\\emph{incorrect}} assertion. A $\\Sigma$-protocol \\dots?\n\nOptions:\nA. has special soundness.\nB. is zero-knowledge.\nC. is a 3-move interaction.\nD. has the verifier polynomially bounded.\n\nAnswer:", | |
| "Question: How does matrix factorization address the issue of missing ratings?\n?\n\nOptions:\nA. It uses regularization of the rating matrix\nB. It performs gradient descent only for existing ratings\nC. It sets missing ratings to zero\nD. It maps ratings into a lower-dimensional space\n\nAnswer:", | |
| "Question: Which of the following statements about coverage-guided fuzzing is/are correct?\n\nOptions:\nA. Fuzzers that have higher code coverage always find more\nB. seed can have a significant impact on further exploration.\nC. Due to the coverage feedback, a small random perturbation of a\nD. Counting the number of times the covered code has been executed\n\nAnswer:", | |
| "Question: Tick the \\textbf{false} assertion.\n\nOptions:\nA. Black-box ZK (zero knowledge) is a stronger notion than (simple) ZK.\nB. We can give a black-box ZK protocol deciding 3-COL (coloring graphs with 3 colours).\nC. The NP language has no ZK proofs.\nD. We can give a ZK protocol deciding ISO (graph isomorphisms).\n\nAnswer:", | |
| "Question: Tick the \\emph{correct} statement. $\\Sigma$-protocols \\ldots?\n\nOptions:\nA. are defined for any language in \\textrm{PSPACE}.\nB. have a polynomially unbounded extractor that can yield a witness.\nC. respect the property of zero-knowledge for any verifier.\nD. consist of protocols between a prover and a verifier, where the verifier is polynomially bounded.\n\nAnswer:", | |
| "Question: Which of the following encryption schemes is deterministic?\n\nOptions:\nA. RSA-OAEP\nB. Plain ElGamal\nC. Plain Rabin\nD. PKCS\\#1.5\n\nAnswer:", | |
| "Question: Let $f_{\\mathrm{MLP}}: \\mathbb{R}^{d} \\rightarrow \\mathbb{R}$ be an $L$-hidden layer multi-layer perceptron (MLP) such that $$ f_{\\mathrm{MLP}}(\\mathbf{x})=\\mathbf{w}^{\\top} \\sigma\\left(\\mathbf{W}_{L} \\sigma\\left(\\mathbf{W}_{L-1} \\ldots \\sigma\\left(\\mathbf{W}_{1} \\mathbf{x}\\right)\\right)\\right) $$ with $\\mathbf{w} \\in \\mathbb{R}^{M}, \\mathbf{W}_{1} \\in \\mathbb{R}^{M \\times d}$ and $\\mathbf{W}_{\\ell} \\in \\mathbb{R}^{M \\times M}$ for $\\ell=2, \\ldots, L$, and $\\sigma$ is an entry-wise activation function. Also, let $f_{\\mathrm{CNN}}: \\mathbb{R}^{d} \\rightarrow \\mathbb{R}$ be an $L^{\\prime}$-hidden layer convolutional neural network (CNN) such that $$ f_{\\mathrm{CNN}}(\\mathbf{x})=\\mathbf{w}^{\\top} \\sigma\\left(\\mathbf{w}_{L^{\\prime}} \\star \\sigma\\left(\\mathbf{w}_{L^{\\prime}-1} \\star \\ldots \\sigma\\left(\\mathbf{w}_{1} \\star \\mathbf{x}\\right)\\right)\\right) $$ with $\\mathbf{w} \\in \\mathbb{R}^{d}, \\mathbf{w}_{\\ell} \\in \\mathbb{R}^{K}$ for $\\ell=1, \\ldots, L^{\\prime}$ and $\\star$ denoting the one-dimensional convolution operator with zero-padding, i.e., output of the convolution has the same dimensionality as the input. Let's assume $\\sigma$ is a tanh activation function. Thus, by flipping the signs of all of the weights leading in and out of a hidden neuron, the input-output mapping function represented by the network is unchanged. Besides, interchanging the values of all of the weights (i.e., by permuting the ordering of the hidden neurons within the layer) also leaves the network input-output mapping function unchanged. Suppose that, given the training data, SGD can find a solution with zero training loss, and the (absolute value) weights of such solution are non-zero and all unique. Choose the largest lower bound on the number of solutions (with zero training loss) achievable by $f_{\\mathrm{MLP}}$ with $L=1$ and $M$ hidden units on this dataset.\n\nOptions:\nA. $M! 2^M$\nB. $1$\nC. $2^M$\nD. $M !$\n\nAnswer:", | |
| "Question: Birthday attacks \\dots?\n\nOptions:\nA. are used to break Google Calendars.\nB. can be used to find collisions in hash functions.\nC. are equivalent to exhaustive search.\nD. imply that a majority of people is born in Spring.\n\nAnswer:", | |
| "Question: Which of the following tasks would typically not be solved by clustering?\n\nOptions:\nA. Community detection in social networks\nB. Discretization of continuous features\nC. Spam detection in an email system\nD. Detection of latent topics in a document collection\n\nAnswer:", | |
| "Question: Which of the following statements about fuzzing is/are correct?\n\nOptions:\nA. generator policies) than mutational fuzzing, but can generate\nB. Blackbox fuzzers can make use of initial seeds.\nC. Generational fuzzing requires more manual work (to specify the\nD. high-quality seeds.\n\nAnswer:", | |
| "Question: Which of these plaintexts can be the result of the decryption of \\texttt{SERGEV} using a simple subsitution cipher?\n\nOptions:\nA. \\texttt{VERNAM}\nB. \\texttt{ENIGMA}\nC. \\texttt{TURING}\nD. \\texttt{CAESAR}\n\nAnswer:", | |
| "Question: Tick the \\textbf{\\emph{incorrect}} assertion. In an interactive proof system for a language $L$, having zero-knowledge implies that $\\ldots$?\n\nOptions:\nA. $\\exists$ ppt $\\mathcal{S}$ such that, for any $\\mathbf{x \\in L}$, $\\mathcal{S}(x,r)$ produces an output indistinguishable from the view of a \\textbf{honest verifier} interacting with a \\textbf{honest prover}.\nB. $\\exists$ ppt $\\mathcal{S}$ such that, for any $\\mathbf{x \\in L}$, $\\mathcal{S}(x,r)$ produces an output indistinguishable from the view of a \\textbf{malicious verifier} interacting with a \\textbf{honest prover}.\nC. $\\exists$ ppt $\\mathcal{S}$ such that, for any $\\mathbf{x \\not\\in L}$, $\\mathcal{S}(x,r)$ produces an output indistinguishable from the view of a \\textbf{honest verifier} interacting with a \\textbf{malicious prover}.\nD. the prover proves the membership of $x$ to $L$ without revealing any secret to the verifier.\n\nAnswer:", | |
| "Question: In a FP tree, the leaf nodes are the ones with:?\n\nOptions:\nA. Lowest confidence\nB. Lowest support\nC. Least in the alphabetical order\nD. None of the above\n\nAnswer:", | |
| "Question: Select all true statements.A penalty will be applied for any incorrect answers.\n\nOptions:\nA. In mapping methods used for visualization, the target space is considered a sub-space of the original space.\nB. The k-NN algorithm is a non-hierarchical, non-overlapping clustering method.\nC. The k-means algorithm always converges into a global minimum.\nD. The k-means algorithm always converges because at each step it minimizes the intra-class variance.\n\nAnswer:", | |
| "Question: Which of the following is wrong regarding Ontologies?\n\nOptions:\nA. We can create more than one ontology that conceptualize the same real-world entities\nB. Ontologies help in the integration of data expressed in different models\nC. Ontologies support domain-specific vocabularies\nD. Ontologies dictate how semi-structured data are serialized\n\nAnswer:", | |
| "Question: Let $n$ be an RSA modulus. Tick the \\emph{false} assertion.\n\nOptions:\nA. The knowledge of $\\lambda(n)$ allows to factorize $n$.\nB. The knowledge of $\\lambda(n)$ allows to recover the RSA secret exponent.\nC. The knowledge of $\\lambda(n)$ allows to decrypt any ciphertext encrypted with the public exponent.\nD. The knowledge of $\\lambda(n)$ allows to factorize $\\lambda(n)$.\n\nAnswer:", | |
| "Question: Tick the most accurate answer. Generating an $\\ell$-bit prime number can be done in?\n\nOptions:\nA. $O(\\ell^2)$ time.\nB. $O(\\ell^4)$ time.\nC. $O(\\ell^3)$ time.\nD. $O(\\ell)$ time.\n\nAnswer:", | |
| "Question: Select valid answers about file descriptors (FD):?\n\nOptions:\nA. The value of FD is unique for every file in the operating system.\nB. FD is usually used as an argument for read and write.\nC. FD is constructed by hashing the filename.\nD. FDs are preserved after fork() and can be used in the new process pointing to the original files.\n\nAnswer:", | |
| "Question: Given this program snippet which is part of a large (> 10000 LoC) codebase,\n which of these statements are true, given that the contents of string \"s\"\n are attacker controlled, the attacker can run the function f only once, the\n attacker has access to the binary and the binary is compiled for x86\\_64 on\n a modern Linux system?\n\nOptions:\nA. and no other mitigation, an attacker can gain remote code execution.\nB. If this program is compiled with DEP (Data-Execution Prevention)\nC. If this program is compiled with no mitigations, an attacker can\nD. gain remote code execution.\n\nAnswer:", | |
| "Question: Tick the \\emph{false} assertion. Let $X$ be the plaintext, $Y$ be the ciphertext and $K$ be the key.\n\nOptions:\nA. The scheme is perfectly secure iff $X$ is independent of $Y$.\nB. The scheme is perfectly secure iff for all $x,y$ $ \\Pr[Y=y] \\neq 0 \\Rightarrow \\Pr [X=x | Y =y ] = \\Pr[X=x]\\;.$\nC. It is possible to achieve perfect secrecy with $K$ smaller than $X$.\nD. Perfect secrecy is too expensive for most of the applications.\n\nAnswer:", | |
| "Question: What is the benefit of LDA over LSI?\n\nOptions:\nA. LSI is sensitive to the ordering of the words in a document, whereas LDA is not\nB. LDA has better theoretical explanation, and its empirical results are in general better than LSI\u2019s\nC. LSI is based on a model of how documents are generated, whereas LDA is not\nD. LDA represents semantic dimensions (topics, concepts) as weighted combinations of terms, whereas LSI does not\n\nAnswer:", | |
| "Question: The following member of the exponential family represents a scalar Gaussian: $p(y)=\\exp \\left\\{(2,-1)\\left(y, y^{2}\\right)^{\\top}-\\right.$ $\\left.1-\\frac{1}{2} \\ln (\\pi)\\right\\}$. What are the mean $\\mu$ and the variance $\\sigma^{2}$ ?\n\nOptions:\nA. (c) $\\mu=1, \\sigma^{2}=0$.\nB. (d) $\\mu=-1, \\sigma^{2}=\\frac{1}{2}$\nC. (a) $\\mu=-1, \\sigma^{2}=0$.\nD. (b) $\\mu=0, \\sigma^{2}=0$.\n\nAnswer:", | |
| "Question: Which of the following statements regarding topic models is false?\n\nOptions:\nA. Topic models map documents to dense vectors\nB. In LDA, topics are modeled as distributions over documents\nC. LDA assumes that each document is generated from a mixture of topics with a probability distribution\nD. Topics can serve as features for document classification\n\nAnswer:", | |
| "Question: Which one of these is \\emph{not} a hard computational problem?\n\nOptions:\nA. Factoring.\nB. Extracting square roots.\nC. Computing the Jacobi symbol.\nD. Computing the discrete log.\n\nAnswer:", | |
| "Question: Tick the \\textbf{incorrect} assertion. A random oracle\\dots?\n\nOptions:\nA. returns a value which is uniformly distributed and independent from the previous queries.\nB. is replaced by a hash function in practice.\nC. is a useful tool in theory.\nD. has unpredictable outputs before queries are made.\n\nAnswer:", | |
| "Question: When constructing a word embedding, what is true regarding negative samples?\n\nOptions:\nA. They are words that do not appear as context words\nB. They are selected among words which are not stop words\nC. Their frequency is decreased down to its logarithm\nD. They are oversampled if less frequent\n\nAnswer:", | |
| "Question: What is the gradient of $\\mathbf{x}^{\\top} \\mathbf{W} \\mathbf{x}$ with respect to $\\mathbf{x}$ (written as a vector)?\n\nOptions:\nA. (b) $\\mathbf{W}^{\\top} \\mathbf{x}$\nB. (c) $\\square\\left(\\mathbf{W}+\\mathbf{W}^{\\top}\\right) \\mathbf{x}$\nC. (a) $\\mathbf{W} \\mathbf{x}$\nD. (d) $\\mathbf{W}$\n\nAnswer:", | |
| "Question: Why is it not recommended to use plain RSA in practice?\n\nOptions:\nA. because of quantum computers.\nB. because of the length of the public keys.\nC. because of the trapdoor one-wayness of the RSA function.\nD. because of the homomorphic property of the RSA function.\n\nAnswer:", | |
| "Question: Tick the \\textbf{false} statement. Let $p$ be a prime number, ...\n\nOptions:\nA. $\\forall x \\in \\mathbb{Z}_p$, $x \\neq 0 \\Longrightarrow x$ is a generator of $\\mathbb{Z}_p$.\nB. $\\forall x \\in \\mathbb{Z}_p^*$, $x \\neq 0 \\Longrightarrow x$ is a generator of $\\mathbb{Z}_p^*$.\nC. $\\forall x \\in \\mathbb{Z}_p^*$, $x$ is invertible.\nD. $\\mathbb{Z}_p^*$ is isomorphic to $\\mathbb{Z}_{p-1}$.\n\nAnswer:", | |
| "Question: Tick the \\emph{incorrect} assertion regarding the security of the Diffie-Hellman key exchange over a subgroup $\\langle g \\rangle \\subset \\mathbb{Z}_p^*$.\n\nOptions:\nA. $\\langle g \\rangle$ should have prime order.\nB. We must ensure that $X\\in \\langle g \\rangle$ for every received $X$.\nC. The binary representation of the output of the key exchange is a uniformly distributed bitstring.\nD. We must ensure that $X\\neq1$ for every received $X$.\n\nAnswer:", | |
| "Question: Which statement about Singular Value Decomposition (SVD) is true?\n\nOptions:\nA. The original matrix M can be reconstructed from the matrices K and D computed by the SVD\nB. SVD, after eliminating the largest singular values, can be interpreted as a least square approximation to the original matrix\nC. The singular values can be computed from the Eigenvalues of the original matrix M\nD. SVD can be computed by constructing Eigenvectors of matrices derived from the original matrix M\n\nAnswer:", | |
| "Question: Which one of these is \\emph{not} a stream cipher?\n\nOptions:\nA. IDEA\nB. RC4\nC. A5/1\nD. E0\n\nAnswer:", | |
| "Question: Which of the following statements has been proved.\n\nOptions:\nA. $\\textrm{IP} \\subseteq \\textrm{NP}$\nB. $\\textrm{IP} = \\textrm{PSPACE}$\nC. $\\textrm{P} \\neq \\textrm{NP}$\nD. $\\textrm{SAT} \\in \\textrm{P}$\n\nAnswer:", | |
| "Question: What is the content of the superblock in the JOS file system?\n\nOptions:\nA. Node with the root directory ('/')\nB. List of all inodes\nC. Total number of blocks on disk\nD. Magic number identifying the file system\n\nAnswer:", | |
| "Question: For this question, one or more assertions can be correct. Tick only the correct assertion(s). There\nwill be a penalty for wrong assertions ticked.Using a 3-gram character model, which of the following expressions are equal to \\( P(\\text{opossum}) \\) ?\n\nOptions:\nA. \\( P(\\text{opo}) \\cdot P(\\text{ssu}) \\cdot P(\\text{sum}) \\)\nB. \\( P(\\text{opo}) \\cdot P(\\text{pos}) \\cdot P(\\text{oss}) \\cdot P(\\text{ssu}) \\cdot P(\\text{sum}) \\)\nC. \\( P(\\text{o}) \\cdot P(\\text{p}|\\text{o}) \\cdot P(\\text{o}|\\text{p}) \\cdot P(\\text{s}|\\text{o}) \\cdot P(\\text{s}|\\text{s}) \\cdot P(\\text{u}|\\text{s}) \\cdot P(\\text{m}|\\text{u}) \\)\nD. \\( P(\\text{opo}) \\cdot P(\\text{s}|\\text{po}) \\cdot P(\\text{s}|\\text{os}) \\cdot P(\\text{u}|\\text{ss}) \\cdot P(\\text{m}|\\text{su}) \\)\n\nAnswer:", | |
| "Question: Which of the following is/are true about fuzzing?\n\nOptions:\nA. Fuzzing can only be applied to C/C++ programs.\nB. Fuzzing open-source software allows the analyst to modify the\nC. When fuzzing open-source software, recompiling it with\nD. mitigations disabled will improve the fuzzing process.\n\nAnswer:", | |
| "Question: The worst case complexity of an exaustive search against DES is\\dots?\n\nOptions:\nA. $1$\nB. $\\frac{2^{64}}{2}$\nC. $2^{56}$\nD. $2^{64}$\n\nAnswer:", | |
| "Question: Which of the following operations would switch the user program from user space to kernel space?\n\nOptions:\nA. Dividing integer by 0.\nB. Calling sin() in math library.\nC. Invoking read() syscall.\nD. Jumping to an invalid address.\n\nAnswer:", | |
| "Question: We apply a Gaussian Mixture Model made of $K$ isotropic Gaussians (invariant to rotation around its center) to $N$ vectors of dimension $D$. What is the number of \\emph{free} parameters of this model?\n ?\n\nOptions:\nA. $2NKD$\nB. $NKD + NKD^2 + N$\nC. $2NKD + N$\nD. $KD + K - 1$\n\nAnswer:", | |
| "Question: Daemons are just long running processes. When applying mitigations to these\n processes, several aspects change. Which ones?\n\nOptions:\nA. different users are handled in a single process.\nB. turned on, allowing the attacker to inject new code.\nC. CFI becomes less effective as the concurrent clients cause\nD. DEP becomes less effective as compiler optimizations are\n\nAnswer:", | |
| "Question: Regarding communities, which of the following is true?\n\nOptions:\nA. Modularity is a measure of how communities are connected together\nB. Agglomerative algorithms recursively decompose communities into sub-communities\nC. Divisive algorithms are based on modularity\nD. Girvan-Newman works by removing edges with the highest betweenness measure\n\nAnswer:", | |
| "Question: A posting indicates?\n\nOptions:\nA. The frequency of a term in the vocabulary\nB. The frequency of a term in a document\nC. The occurrence of a term in a document\nD. The list of terms occurring in a document\n\nAnswer:", | |
| "Question: Pick the \\textit{correct} assertion. Kasiski test can be used to \\dots?\n\nOptions:\nA. distinguish good students in cryptography classes.\nB. break a cipher that provides perfect secrecy.\nC. break the ROT13 cipher.\nD. break the Vigen\\`ere cipher.\n\nAnswer:", | |
| "Question: Tick the \\textbf{false} statement regarding Modes of Operation in Block ciphers.\n\nOptions:\nA. The CBC mode uses an IV value.\nB. The ECB mode is a secure way to use block ciphers using a single key.\nC. The CTR mode uses a counter which is new for every block.\nD. The OFB mode transforms a block cipher into a stream cipher.\n\nAnswer:", | |
| "Question: Assume we run the Louvain algorithm to find communities in the following graph. Let\u00a0\u0394Q(\ud835\udc56 \u2192 \ud835\udc34) and \u0394Q(\ud835\udc56 \u2192 \ud835\udc35) denote the modularity gain of joining node \ud835\udc56 to community\u00a0\ud835\udc34 and \ud835\udc35 respectively. Which is true?\n\nOptions:\nA. \u0394Q(\ud835\udc56\u2192\ud835\udc34)>\u0394Q(\ud835\udc56\u2192\ud835\udc35)\nB. \u0394Q(\ud835\udc56\u2192\ud835\udc34)=\u0394Q(\ud835\udc56\u2192\ud835\udc35)\nC. \u0394Q(\ud835\udc56\u2192\ud835\udc34)<\u0394Q(\ud835\udc56\u2192\ud835\udc35)\nD. All options are possible.\n\nAnswer:", | |
| "Question: One can find a collision in a hash function $h\\colon \\{0,1\\}^* \\rightarrow \\{0,1\\}^n$ with expected time complexity\\dots?\n\nOptions:\nA. $\\Theta(\\sqrt{n})$.\nB. $\\Theta(n)$.\nC. $\\Theta(2^n)$.\nD. $\\Theta(2^{n/2})$.\n\nAnswer:", | |
| "Question: Assume a user program executes following tasks. Select all options that will use a system call.\n\nOptions:\nA. Read the user's input \"Hello world\" from the keyboard.\nB. Write \"Hello world\" to a file.\nC. Encrypt \"Hello world\" by AES.\nD. Send \"Hello world\" to another machine via Network Interface Card.\n\nAnswer:", | |
| "Question: Which one of these Ciphers is perfectly secure?\n\nOptions:\nA. Hieroglyphs\nB. Caesar\nC. Vigen\\`{e}re\nD. Vernam\n\nAnswer:", | |
| "Question: Tick the \\textbf{false} statement. The Shannon Encryption Model ...\n\nOptions:\nA. requires a black-box encryption model.\nB. assumes a known input distribution.\nC. assumes the key is independent from the message.\nD. requires the correctness property $\\Pr[C_K^{-1}(C_K(X))=X]=1$.\n\nAnswer:", | |
| "Question: Given the following teleporting matrix (\u0395) for nodes A, B and C:[0\u00a0\u00a0\u00a0 \u00bd\u00a0\u00a0\u00a0 0][0\u00a0\u00a0\u00a0\u00a0 0\u00a0\u00a0\u00a0 0][0\u00a0\u00a0\u00a0 \u00bd\u00a0\u00a0\u00a0 1]and making no assumptions about the link matrix (R), which of the following is correct:(Reminder: columns are the probabilities to leave the respective node.)?\n\nOptions:\nA. A random walker can never reach node A\nB. A random walker can never leave node A\nC. A random walker can always leave node C\nD. A random walker can always leave node B\n\nAnswer:", | |
| "Question: Which one is \\emph{not} a security requirement on hash functions?\n\nOptions:\nA. onewayness\nB. bijectivity\nC. pseudo-randomness\nD. collision resistance\n\nAnswer:", | |
| "Question: Let $C_1$, $C_2$ and $C^*$ be three independent random permutations over a set $A$, $C^*$ being uniformaly distributed. Tick the \\textbf{false} assertion.\n\nOptions:\nA. $[C_2\\circ C_1]^n=[C_1]^n\\circ [C_2]^n$\nB. $[C^*]^n\\times [C_2]^n=[C^*]^n$\nC. $[C_1]^n\\times [C^*]^n=[C^*]^n$\nD. $[C^*]^n$ is neutral for $x$\n\nAnswer:", | |
| "Question: Let $E_{a,b}(\\mathrm{GF}(p))$ be non-singular elliptic curve with prime order, with $p$ a 256-bit prime and let $P,Q,R \\in E_{a,b}(\\mathrm{GF}(p)) \\backslash \\{\\mathcal{O}\\}$ such that $R=aP$ for an integer $a > 0$. Tick the \\textit{correct} assertion.\n\nOptions:\nA. It is hard to subtract $P$ from $Q$.\nB. The point at infinity $\\mathcal{O}$ does not have any inverse point.\nC. Given $P$ and $R$, it is hard to recover $a$.\nD. To compute the point $P+Q$, we just have to compute $(x_P+x_Q \\bmod{p}, y_P+y_Q \\bmod{p})$.\n\nAnswer:", | |
| "Question: Tick the \\textbf{false} assertion. In order to have zero-knowledge from $\\Sigma$-protocols, we need to add the use of \\ldots?\n\nOptions:\nA. \\ldots an ephemeral key $h$ and a Pedersen commitment.\nB. \\ldots a common reference string.\nC. \\ldots hash functions.\nD. \\ldots none of the above is necessary, zero-knowledge is already contained in $\\Sigma$-protocols.\n\nAnswer:", | |
| "Question: What is a good distance metric to be used when you want to compute the similarity between documents independent of their length?A penalty will be applied for any incorrect answers.\n\nOptions:\nA. Cosine similarity\nB. Euclidean distance\nC. Manhattan distance\nD. Chi-squared distance\n\nAnswer:", | |
| "Question: Let $X_1,X_2,\\dots$ be i.i.d. binary random variables with $p_{X_i}(1) =\frac{1}{4}$ for every $i\\geq 1$. Let $Y_1$ be a uniform binary random variable, and let $Y_i = Y_{i-1} \\oplus X_{i-1}$ for every $i\\geq 2$, where $\\oplus$ denotes the modulo-2 sum. For any given $n\\geq 1$, what is the value of $H(Y_1,Y_2,\\dots,Y_n)$? [Hint: what is the value of $H(Y_i|Y_1,\\dots,Y_{i-1})$?]?\n\nOptions:\nA. $\\left(2-\frac{3}{4}\\log 3\right) n + \frac{3}{4}\\log 3 - 1$.\nB. $n$.\nC. $\\left(2-\frac{3}{4}\\log 3\right) n + 1$.\nD. $\\left(3 - \frac{3}{4}\\log 3\right) n +\frac{3}{4}\\log 3 -2$.\n\nAnswer:", | |
| "Question: Which of the following is/are true about fuzzing?\n\nOptions:\nA. inputs that comply with all the format rules.\nB. The quality of initial seeds matters in mutational fuzzing.\nC. Black box fuzzing may struggle to find inputs that reach deep into the program.\nD. In structure-aware fuzzing, the mutator should only generate\n\nAnswer:", | |
| "Question: Tick the \\textbf{true} assertion. Assume that $p$ is prime.\n\nOptions:\nA. $QR(p)$ is of order $\\frac{p-1}{4}$\nB. $\\mathbf{Z}_{p}^*$ has only one generator.\nC. $\\mathbf{Z}_{p}^*$ has $\\varphi(\\varphi(p))$ generators.\nD. All elements of $\\mathbf{Z}_{p}$ are invertible.\n\nAnswer:", | |
| "Question: Which of the following apply to recent Android-based mobile systems but not\n to Linux-based desktop systems?\n\nOptions:\nA. Apps should use the binder interface to communicate with other\nB. By default, each app runs as its own user.\nC. apps.\nD. All apps run in a strict container with only limited system\n\nAnswer:", | |
| "Question: Which of the following is correct regarding prediction models?\n\nOptions:\nA. A high bias is a sign of overfitting.\nB. A high variance is a sign of underfitting.\nC. In low data regime, complex models tend to perform better.\nD. Simple models have higher bias than complex models.\n\nAnswer:", | |
| "Question: Let $n$ be any positive integer. Three of the following assertions are equivalent. Tick the remaining one.\n\nOptions:\nA. $\\mathbb{Z}_n$ is a field.\nB. $\\varphi(n)=n-1 $, where $\\varphi$ denotes the Euler totient function.\nC. $n$ is a prime power.\nD. Any element $x \\in \\mathbb{Z}_n \\backslash \\{0\\}$ is invertible.\n\nAnswer:", | |
| "Question: Consider a binary classification problem with classifier $f(\\mathbf{x})$ given by $$ f(\\mathbf{x})= \\begin{cases}1, & g(\\mathbf{x}) \\geq 0 \\\\ -1, & g(\\mathbf{x})<0\\end{cases} $$ and $\\mathbf{x} \\in \\mathbb{R}^{6}$. Consider a specific pair $(\\mathbf{x}, y=1)$ and assume that $g(\\mathbf{x})=8$. In particular this means that this point is classified correctly by $f$. Assume further that we have computed the gradient of $g$ at $\\mathbf{x}$ to be $\\nabla_{\\mathbf{x}} g(\\mathbf{x})=(+1,-2,+3,-4,+5,-6)$. You are allowed to make one step in order to (hopefully) find an adversarial example. In the following four questions, assume $\\epsilon=1$. What is the value of $g(\\mathbf{x}+\\delta)$ for this $\\ell_{\\infty}$-optimal choice assuming that $g$ is (locally) linear?\n\nOptions:\nA. $2$\nB. $-4$\nC. $+13$\nD. $-13$\n\nAnswer:", | |
| "Question: For a blockcipher $B:\\{0,1\\}^k\\times \\{0,1\\}^n \\rightarrow \\{0,1\\}^n$ that has decorrelation $Dec^q_{\\| \\cdot \\|_{\\infty}}(B,C^*)=d$ (from a perfect cipher $C^*$), the best advantage of \\textit{any} distinguisher that makes $q$ queries is \\ldots?\n\nOptions:\nA. bounded by $d/2$.\nB. not related to $d$; we have to use the $a$-norm to get a more general result.\nC. bounded by $d$.\nD. bounded by $d-\\frac{1}{2}$.\n\nAnswer:", | |
| "Question: Tick the \\emph{correct} assertion. In ElGamal $\\ldots$?\n\nOptions:\nA. the encryption algorithm is deterministic.\nB. the key recovery problem is equivalent to the Computational Diffie Hellman problem.\nC. the size of the ciphertext is always bigger than the size of the corresponding plaintext.\nD. the decryption problem can be hard even if the discrete logarithm is easy to compute in the underlying group.\n\nAnswer:", | |
| "Question: How many $x \\in \\mathbb{Z} / 23\\mathbb{Z}$ satisfy the equation $0 = 1 - x + x^2 - x^3 + .... - x^{21} + x^{22} - x^{23}$, when all operations are with respect to the field $(\\mathbb{Z} / 23 \\mathbb{Z}, + , \\cdot)$? Check the correct answer.\n\nOptions:\nA. 22\nB. 2\nC. 0\nD. 1\n\nAnswer:", | |
| "Question: Which of the following statements are true?\n\nOptions:\nA. The more training examples, the more accurate the prediction of a $k$-nearest-neighbor classifier.\nB. k-nearest-neighbors cannot be used for regression.\nC. A $k$-nearest-neighbor classifier is sensitive to outliers.\nD. Training a $k$-nearest-neighbor classifier takes more computational time than applying it / using it for prediction.\n\nAnswer:", | |
| "Question: Let $n$ be an integer. Tick the \\emph{true} assertion about the Miller-Rabin Primality Test.\n\nOptions:\nA. If the algorithms outputs $prime$, then $n$ is definitely a prime.\nB. If the algorithms outputs $composite$, then $n$ is definitely \\emph{not} a prime.\nC. The test can be used to factorize $n$ if it is composite.\nD. The test always outputs $prime$ if $n$ is a Carmichael number.\n\nAnswer:", | |
| "Question: Consider:Non-terminals: S (top-level), NP (for \"noun phrase\"), VP (for \"verbal phrase\"), N (for \"Noun\"), V (for \"Verb\"), Det (for \"Determiner\").PoS tags: N, V, DetTerminals: I, yesterday, in, rain, went, home, the, cat, goOut of the following, select the ones which are possible valid \"syntactic rules\" as defined in a context-free grammar for processing (a tiny part of) English.A penalty will be applied for any incorrect answers.\n\nOptions:\nA. NP \u2192 N\nB. S\u00a0\u2192 NP VP\nC. V \u2192 VP N\u00a0\nD. NP \u2192 Det N\n\nAnswer:", | |
| "Question: What is the support of the itemset {beer, chocolate} and the confidence of the rule\u00a0{chocolate} \u2192 {beer} in the dataset below?TID: Items BoughtT1: milk, butter, beerT2: milk, diaper, beer, eggsT3: butter, diaper, beer, chocolate\nT4: milk, butter, diaper, beerT5: milk, butter, diaper, chocolate\n\n\n\n?\n\nOptions:\nA. 0.2/0.25\nB. 0.4/0.5\nC. 0.2/0.5\nD. 0.4/0.25\n\n\n\n\n\nAnswer:", | |
| "Question: In an interactive proof system for a language $L$, having $\\beta$-soundness means that\\dots?\n\nOptions:\nA. if we run the protocol with input $x\\not\\in L$, with a \\textbf{malicious prover}, and a \\textbf{honest verifier} the probability that the protocol succeeds is upper-bounded by $\\beta$.\nB. if we run the protocol with input $x\\in L$, with a \\textbf{malicious prover}, and a \\textbf{honest verifier} the probability that the protocol succeeds is upper-bounded by $\\beta$.\nC. if we run the protocol with input $x\\in L$, with a \\textbf{honest prover}, and a \\textbf{malicious verifier} the probability that the protocol succeeds is upper-bounded by $\\beta$.\nD. if we run the protocol with input $x\\in L$, with a \\textbf{honest prover}, and a \\textbf{honest verifier} the probability that the protocol succeeds is upper-bounded by $\\beta$.\n\nAnswer:", | |
| "Question: Which of the following properties is part of the RDF Schema Language?\n\nOptions:\nA. Type\nB. Predicate\nC. Description\nD. Domain\n\nAnswer:", | |
| "Question: Which of the following attack vectors apply to mobile Android systems?\n\nOptions:\nA. from benign apps.\nB. malicious apps to steal access to their privileges.\nC. Apps may maliciously declare intent filters to receive intents\nD. Overprivileged apps may be abused as a confused deputy, allowing\n\nAnswer:", | |
| "Question: Tick the \\emph{incorrect} assertion. In a multiplicative cyclic group $G$ of order $m > 1$ with neutral element $e_G$ \\ldots?\n\nOptions:\nA. $\\lambda = m$, where $\\lambda$ is the exponent of $G$.\nB. the order of every element $x \\in G$ is $m$.\nC. there exists $g \\in G$ that generates the whole group.\nD. for any $x \\in G$, we have that $x^m = e_{G}$.\n\nAnswer:", | |
| "Question: Select the \\emph{weakest} algorithm.\n\nOptions:\nA. A5/4\nB. A5/2\nC. A5/3\nD. A5/1\n\nAnswer:", | |
| "Question: Using a salt in a password authentication protocol\\dots?\n\nOptions:\nA. adds more spice in the protocol.\nB. avoids single-target and multi-target bruteforce attacks.\nC. \\textbf{requires} the salt to be encrypted in the database.\nD. avoids multi-target but not single-target bruteforce attacks.\n\nAnswer:", | |
| "Question: Select the \\emph{incorrect} statement. The discrete logarithm?\n\nOptions:\nA. can be solved by a polynomial algorithm.\nB. is an easy problem in the Abelian group $\\mathbb{Z}_p$.\nC. is a hard problem in the multiplicative group $\\mathbb{Z}_p^*$.\nD. can be solved easily on a quantum computer.\n\nAnswer:", | |
| "Question: Select all statements that are true.A penalty will be applied for any wrong answers.\n\nOptions:\nA. The recognizer functionality of a parser decides if a given sequence of words is syntactically correct or not.\nB. Determining whether a sentence has a pragmatic meaning depends on the context that is available.\nC. For a sentence to be acceptable in general, it is sufficient to satisfy the positional and selectional constraints of a given language.\nD. The analyzer functionality of a parser determines the set of all possible associated syntactic structures for any syntactically correct sentence.\n\nAnswer:", | |
| "Question: Which of the following arguments is correct? ?\n\nOptions:\nA. All students in this class understand math. Alice is a student in this class. Therefore, Alice doesn\u2019t understand math.\nB. Every physics major takes calculus. Mathilde is taking calculus. Therefore, Mathilde is a physics major.\nC. All cats like milk. My pet is not a cat. Therefore, my pet does not like milk.\nD. Everyone who eats vegetables every day is healthy. Linda is not healthy. Therefore, Linda does not eat vegetables every day.\n\nAnswer:", | |
| "Question: Select all valid answers about UNIX-like shell.\n\nOptions:\nA. The shell is the layer, which has to be always used for communicating with kernel.\nB. The shell must run only in a single instance. Multiple running instances cause memory corruption.\nC. The shell is a user interface for UNIX-like systems.\nD. The shell is a function inside kernel.\n\nAnswer:", | |
| "Question: Select the statements that are true.A penalty will be applied to any incorrect answers selected.\n\nOptions:\nA. The vector space model represents documents as vectors derived from the distribution of indexing terms in the document.\nB. Information retrieval is the selection of documents relevant to a query from an unstructured collection of documents.\nC. Different IR systems can differ in the way they represent documents, represent queries, and define the relevance measure between documents and queries.\nD. The dimensionality of the vector space does not depend on the size of the indexing vocabulary.\n\nAnswer:", | |
| "Question: Tick the \\emph{incorrect} statement. The following statements have been proven:?\n\nOptions:\nA. $\\textrm{SAT} \\in \\textrm{PSPACE} \\cap \\textrm{P}$\nB. $\\textrm{P} \\subseteq \\textrm{NP} \\subseteq \\textrm{IP}$\nC. $\\textrm{P} \\subseteq \\textrm{IP} \\cap \\textrm{NP}$\nD. $\\textrm{co-NP} \\subseteq \\textrm{PSPACE}$\n\nAnswer:", | |
| "Question: Which of the following is correct regarding community detection?\n\nOptions:\nA. High betweenness of an edge indicates that the communities are well connected by that edge\nB. The Louvain algorithm attempts to minimize the overall modularity measure of a community graph\nC. High modularity of a community indicates a large difference between the number of edges of the community and the number of edges of a null model\nD. The Girvan-Newman algorithm attempts to maximize the overall betweenness measure of a community graph\n\nAnswer:", | |
| "Question: Assume we enforce CFI for function returns. Which of the following statements are true?\n\nOptions:\nA. CFI on returns will make control-flow hijacking harder\nB. CFI on returns is too coarse-grained and may give the adversary sufficient valid targets for an exploit\nC. CFI on returns ensures that only the single valid target is allowed\nD. CFI on returns cannot support exceptions\n\nAnswer:", | |
| "Question: We want to generate a $\\ell$-bit prime. The complexity is roughly\\dots?\n\nOptions:\nA. $\\ell^2$\nB. $\\ell^3$\nC. $\\ell^4$\nD. $\\ell^5$\n\nAnswer:", | |
| "B", | |
| "B", | |
| "B", | |
| "C", | |
| "B", | |
| "D", | |
| "B", | |
| "C", | |
| "D", | |
| "C", | |
| "D", | |
| "D", | |
| "C", | |
| "A", | |
| "C", | |
| "C", | |
| "A", | |
| "A", | |
| "D", | |
| "A", | |
| "B", | |
| "A", | |
| "B", | |
| "B", | |
| "A", | |
| "A", | |
| "C", | |
| "C", | |
| "C", | |
| "C", | |
| "B", | |
| "D", | |
| "D", | |
| "C", | |
| "C", | |
| "D", | |
| "A", | |
| "C", | |
| "C", | |
| "D", | |
| "A", | |
| "D", | |
| "B", | |
| "A", | |
| "D", | |
| "A", | |
| "C", | |
| "D", | |
| "B", | |
| "D", | |
| "A", | |
| "C", | |
| "A", | |
| "D", | |
| "C", | |
| "D", | |
| "C", | |
| "C", | |
| "D", | |
| "C", | |
| "B", | |
| "B", | |
| "D", | |
| "C", | |
| "A", | |
| "A", | |
| "C", | |
| "B", | |
| "D", | |
| "D", | |
| "A", | |
| "A", | |
| "A", | |
| "C", | |
| "B", | |
| "B", | |
| "A", | |
| "D", | |
| "D", | |
| "B", | |
| "B", | |
| "A", | |
| "B", | |
| "B", | |
| "C", | |
| "B", | |
| "A", | |
| "D", | |
| "B", | |
| "D", | |
| "D", | |
| "B", | |
| "C", | |
| "A", | |
| "D", | |
| "A", | |
| "A", | |
| "C", | |
| "B", | |
| "A", | |
| "D", | |
| "A", | |
| "A", | |
| "D", | |
| "D", | |
| "D", | |
| "B", | |
| "D", | |
| "B", | |
| "A", | |
| "D", | |
| "C", | |
| "A", | |
| "C", | |
| "B", | |
| "C", | |
| "A", | |
| "B", | |
| "B", | |
| "A", | |
| "A", | |
| "C", | |
| "B", | |
| "A", | |
| "D", | |
| "D", | |
| "D", | |
| "A", | |
| "B", | |
| "C", | |
| "A", | |
| "B", | |
| "A", | |
| "A", | |
| "C", | |
| "D", | |
| "B", | |
| "C", | |
| "D", | |
| "C", | |
| "B", | |
| "C", | |
| "C", | |
| "C", | |
| "D", | |
| "B", | |
| "A", | |
| "B", | |
| "B", | |
| "B", | |
| "B", | |
| "D", | |
| "B", | |
| "A", | |
| "C", | |
| "C", | |
| "A", | |
| "B", | |
| "A", | |
| "C", | |
| "D", | |
| "D", | |
| "B", | |
| "B", | |
| "C", | |
| "B", | |
| "B", | |
| "C", | |
| "B", | |
| "A", | |
| "D", | |
| "D", | |
| "A", | |
| "A", | |
| "C", | |
| "A", | |
| "A", | |
| "C", | |
| "B", | |
| "B", | |
| "C", | |
| "C", | |
| "A", | |
| "D", | |
| "D", | |
| "A", | |
| "B", | |
| "C", | |
| "B", | |
| "C", | |
| "A", | |
| "C", | |
| "C", | |
| "C", | |
| "B", | |
| "A", | |
| "D", | |
| "C", | |
| "B", | |
| "D", | |
| "A", | |
| "D", | |
| "B", | |
| "D", | |
| "D", | |
| "C", | |
| "A", | |
| "C", | |
| "C", | |
| "A", | |
| "D", | |
| "C", | |
| "B", | |
| "D", | |
| "C", | |
| "D", | |
| "C", | |
| "B", | |
| "C", | |
| "A", | |
| "A", | |
| "B", | |
| "B", | |
| "D", | |
| "A", | |
| "A", | |
| "A", | |
| "A", | |
| "C", | |
| "C", | |
| "B", | |
| "B", | |
| "B", | |
| "C", | |
| "A", | |
| "B", | |
| "A", | |
| "A", | |
| "B", | |
| "B", | |
| "D", | |
| "A", | |
| "B", | |
| "A", | |
| "B", | |
| "C", | |
| "C", | |
| "B", | |
| "D", | |
| "C", | |
| "A", | |
| "D", | |
| "B", | |
| "A", | |
| "B", | |
| "A", | |
| "B", | |
| "B", | |
| "C", | |
| "B", | |
| "D", | |
| "B", | |
| "B", | |
| "B", | |
| "C", | |
| "B", | |
| "C", | |
| "C", | |
| "B", | |
| "C", | |
| "B", | |
| "D", | |
| "D", | |
| "B", | |
| "C", | |
| "C", | |
| "A", | |
| "B", | |
| "A", | |
| "A", | |
| "B", | |
| "D", | |
| "C", | |
| "B", | |
| "B", | |
| "C", | |
| "A", | |
| "A", | |
| "B", | |
| "A", | |
| "B", | |
| "A", | |
| "A", | |
| "C", | |
| "D", | |
| "B", | |
| "A", | |
| "B", | |
| "D", | |
| "C", | |
| "C", | |
| "A", | |
| "A", | |
| "C", | |
| "D", | |
| "C", | |
| "A", | |
| "D", | |
| "A", | |
| "A", | |
| "C", | |
| "B", | |
| "B", | |
| "A", | |
| "B", | |
| "A", | |
| "C", | |
| "C", | |
| "C", | |
| "C", | |
| "B", | |
| "C", | |
| "D", | |
| "A", | |
| "C", | |
| "A", | |
| "D", | |
| "D", | |
| "C", | |
| "C", | |
| "D", | |
| "A", | |
| "D", | |
| "B", | |
| "A", | |
| "D", | |
| "D", | |
| "A", | |
| "D", | |
| "C", | |
| "C", | |
| "C", | |
| "C", | |
| "B", | |
| "B", | |
| "A", | |
| "B", | |
| "B", | |
| "D", | |
| "A", | |
| "A", | |
| "D", | |
| "B", | |
| "D", | |
| "C", | |
| "B", | |
| "C", | |
| "D", | |
| "D", | |
| "B", | |
| "B", | |
| "A", | |
| "A", | |
| "A", | |
| "B", | |
| "D", | |
| "A", | |
| "D", | |
| "C", | |
| "B", | |
| "B", | |
| "B", | |
| "B", | |
| "A", | |
| "C", | |
| "B", | |
| "A", | |
| "D", | |
| "B", | |
| "A", | |
| "C", | |
| "C", | |
| "B", | |
| "C", | |
| "C", | |
| "B", | |
| "C", | |
| "D", | |
| "D", | |
| "C", | |
| "B", | |
| "C", | |
| "B", | |
| "A", | |
| "C", | |
| "B", | |
| "B", | |
| "D", | |
| "C", | |
| "D", | |
| "D", | |
| "A", | |
| "B", | |
| "B", | |
| "C", | |
| "D", | |
| "A", | |
| "C", | |
| "B", | |
| "A", | |
| "A", | |
| "B", | |
| "D", | |
| "B", | |
| "C", | |
| "B", | |
| "A", | |
| "B", | |
| "D", | |
| "B", | |
| "C", | |
| "D", | |
| "B", | |
| "B", | |
| "A", | |
| "C", | |
| "C", | |
| "A", | |
| "B", | |
| "A", | |
| "A", | |
| "D", | |
| "B", | |
| "D", | |
| "A", | |
| "B", | |
| "A", | |
| "B", | |
| "D", | |
| "C", | |
| "B", | |
| "D", | |
| "C", | |
| "B", | |
| "B", | |
| "D", | |
| "A", | |
| "D", | |
| "A", | |
| "A", | |
| "A", | |
| "A", | |
| "C", | |
| "B", | |
| "B", | |
| "B", | |
| "C", | |
| "C", | |
| "D", | |
| "C", | |
| "C", | |
| "D", | |
| "B", | |
| "A", | |
| "C", | |
| "B", | |
| "B", | |
| "B", | |
| "A", | |
| "C", | |
| "A", | |
| "C", | |
| "B", | |
| "A", | |
| "B", | |
| "A", | |
| "A", | |
| "B", | |
| "C", | |
| "B", | |
| "A", | |
| "B", | |
| "B", | |
| "B", | |
| "D", | |
| "B", | |
| "C", | |
| "C", | |
| "D", | |
| "B", | |
| "A", | |
| "B", | |
| "B", | |
| "C", | |
| "C", | |
| "D", | |
| "B", | |
| "A", | |
| "A", | |
| "A", | |
| "B", | |
| "A", | |
| "B", | |
| "C", | |
| "C", | |
| "C", | |
| "B", | |
| "B", | |
| "C", | |
| "A", | |
| "A", | |
| "C", | |
| "C", | |
| "C", | |
| "A", | |
| "A", | |
| "B", | |
| "C", | |
| "A", | |
| "D", | |
| "B", | |
| "C", | |
| "D", | |
| "A", | |
| "C", | |
| "A", | |
| "C", | |
| "D", | |
| "D", | |
| "D", | |
| "C", | |
| "D", | |
| "A", | |
| "C", | |
| "B", | |
| "B", | |
| "C", | |
| "D", | |
| "C", | |
| "B", | |
| "A", | |
| "D", | |
| "D", | |
| "C", | |
| "D", | |
| "A", | |
| "C", | |
| "D", | |
| "D", | |
| "A", | |
| "B", | |
| "B", | |
| "A", | |
| "B", | |
| "B", | |
| "A", | |
| "D", | |
| "A", | |
| "A", | |
| "C", | |
| "D", | |
| "B", | |
| "B", | |
| "C", | |
| "C", | |
| "D", | |
| "A", | |
| "D", | |
| "B", | |
| "D", | |
| "B", | |
| "D", | |
| "D", | |
| "D", | |
| "D", | |
| "A", | |
| "B", | |
| "B", | |
| "D", | |
| "C", | |
| "D", | |
| "A", | |
| "B", | |
| "B", | |
| "A", | |
| "A", | |
| "D", | |
| "C", | |
| "C", | |
| "A", | |
| "D", | |
| "B", | |
| "C", | |
| "C", | |
| "D", | |
| "B", | |
| "A", | |
| "D", | |
| "C", | |
| "A", | |
| "D", | |
| "B", | |
| "A", | |
| "A", | |
| "D", | |
| "C", | |
| "C", | |
| "C", | |
| "C", | |
| "C", | |
| "C" | |
| ], | |
| "desc_act": false, | |
| "exllama_config": { | |
| "version": 1 | |
| }, | |
| "group_size": 128, | |
| "max_input_length": null, | |
| "model_seqlen": null, | |
| "module_name_preceding_first_block": null, | |
| "modules_in_block_to_quantize": null, | |
| "pad_token_id": null, | |
| "quant_method": "gptq", | |
| "sym": true, | |
| "tokenizer": null, | |
| "true_sequential": true, | |
| "use_cuda_fp16": false, | |
| "use_exllama": true | |
| }, | |
| "rms_norm_eps": 1e-05, | |
| "rope_scaling": null, | |
| "rope_theta": 500000.0, | |
| "tie_word_embeddings": false, | |
| "torch_dtype": "float16", | |
| "transformers_version": "4.40.2", | |
| "use_cache": true, | |
| "vocab_size": 128256 | |
| } | |