id
int64
1
42.8k
arxiv_id
stringlengths
10
13
base_arxiv_id
float64
1.11k
2.51k
version
float64
1
14
title
stringlengths
8
224
abstract
stringlengths
51
4.03k
submit_date
stringdate
2011-09-01 00:00:00
2025-08-13 00:00:00
metadata
stringlengths
283
56.2k
42,807
2110.09022v3
2,110.09022
3
Mitigating Memorization of Noisy Labels via Regularization between Representations
Designing robust loss functions is popular in learning with noisy labels while existing designs did not explicitly consider the overfitting property of deep neural networks (DNNs). As a result, applying these losses may still suffer from overfitting/memorizing noisy labels as training proceeds. In this paper, we first theoretically analyze the memorization effect and show that a lower-capacity model may perform better on noisy datasets. However, it is non-trivial to design a neural network with the best capacity given an arbitrary task. To circumvent this dilemma, instead of changing the model architecture, we decouple DNNs into an encoder followed by a linear classifier and propose to restrict the function space of a DNN by a representation regularizer. Particularly, we require the distance between two self-supervised features to be positively related to the distance between the corresponding two supervised model outputs. Our proposed framework is easily extendable and can incorporate many other robust loss functions to further improve performance. Extensive experiments and theoretical analyses support our claims. Code is available at github.com/UCSC-REAL/SelfSup_NoisyLabel.
2021-10-18
{'id': '2110.09022v3', 'title': 'Mitigating Memorization of Noisy Labels via Regularization between Representations', 'abstract': 'Designing robust loss functions is popular in learning with noisy labels\nwhile existing designs did not explicitly consider the overfitting property of\ndeep neural networks (DNNs). As a result, applying these losses may still\nsuffer from overfitting/memorizing noisy labels as training proceeds. In this\npaper, we first theoretically analyze the memorization effect and show that a\nlower-capacity model may perform better on noisy datasets. However, it is\nnon-trivial to design a neural network with the best capacity given an\narbitrary task. To circumvent this dilemma, instead of changing the model\narchitecture, we decouple DNNs into an encoder followed by a linear classifier\nand propose to restrict the function space of a DNN by a representation\nregularizer. Particularly, we require the distance between two self-supervised\nfeatures to be positively related to the distance between the corresponding two\nsupervised model outputs. Our proposed framework is easily extendable and can\nincorporate many other robust loss functions to further improve performance.\nExtensive experiments and theoretical analyses support our claims. Code is\navailable at github.com/UCSC-REAL/SelfSup_NoisyLabel.', 'authors': ['Hao Cheng', 'Zhaowei Zhu', 'Xing Sun', 'Yang Liu'], 'published': '2021-10-18 05:41:57+00:00', 'categories': ['cs.LG', 'cs.CV'], 'url': 'http://arxiv.org/abs/2110.09022v3'}
42,808
2210.00750v2
2,210.0075
2
Offline Reinforcement Learning with Differentiable Function Approximation is Provably Efficient
Offline reinforcement learning, which aims at optimizing sequential decision-making strategies with historical data, has been extensively applied in real-life applications. State-Of-The-Art algorithms usually leverage powerful function approximators (e.g. neural networks) to alleviate the sample complexity hurdle for better empirical performances. Despite the successes, a more systematic understanding of the statistical complexity for function approximation remains lacking. Towards bridging the gap, we take a step by considering offline reinforcement learning with differentiable function class approximation (DFA). This function class naturally incorporates a wide range of models with nonlinear/nonconvex structures. Most importantly, we show offline RL with differentiable function approximation is provably efficient by analyzing the pessimistic fitted Q-learning (PFQL) algorithm, and our results provide the theoretical basis for understanding a variety of practical heuristics that rely on Fitted Q-Iteration style design. In addition, we further improve our guarantee with a tighter instance-dependent characterization. We hope our work could draw interest in studying reinforcement learning with differentiable function approximation beyond the scope of current research.
2022-10-03
{'id': '2210.00750v2', 'title': 'Offline Reinforcement Learning with Differentiable Function Approximation is Provably Efficient', 'abstract': 'Offline reinforcement learning, which aims at optimizing sequential\ndecision-making strategies with historical data, has been extensively applied\nin real-life applications. State-Of-The-Art algorithms usually leverage\npowerful function approximators (e.g. neural networks) to alleviate the sample\ncomplexity hurdle for better empirical performances. Despite the successes, a\nmore systematic understanding of the statistical complexity for function\napproximation remains lacking. Towards bridging the gap, we take a step by\nconsidering offline reinforcement learning with differentiable function class\napproximation (DFA). This function class naturally incorporates a wide range of\nmodels with nonlinear/nonconvex structures. Most importantly, we show offline\nRL with differentiable function approximation is provably efficient by\nanalyzing the pessimistic fitted Q-learning (PFQL) algorithm, and our results\nprovide the theoretical basis for understanding a variety of practical\nheuristics that rely on Fitted Q-Iteration style design. In addition, we\nfurther improve our guarantee with a tighter instance-dependent\ncharacterization. We hope our work could draw interest in studying\nreinforcement learning with differentiable function approximation beyond the\nscope of current research.', 'authors': ['Ming Yin', 'Mengdi Wang', 'Yu-Xiang Wang'], 'published': '2022-10-03 07:59:42+00:00', 'categories': ['cs.LG', 'cs.AI', 'stat.ML'], 'url': 'http://arxiv.org/abs/2210.00750v2'}
42,809
2303.17264v1
2,303.17264
1
Multifactor Sequential Disentanglement via Structured Koopman Autoencoders
Disentangling complex data to its latent factors of variation is a fundamental task in representation learning. Existing work on sequential disentanglement mostly provides two factor representations, i.e., it separates the data to time-varying and time-invariant factors. In contrast, we consider multifactor disentanglement in which multiple (more than two) semantic disentangled components are generated. Key to our approach is a strong inductive bias where we assume that the underlying dynamics can be represented linearly in the latent space. Under this assumption, it becomes natural to exploit the recently introduced Koopman autoencoder models. However, disentangled representations are not guaranteed in Koopman approaches, and thus we propose a novel spectral loss term which leads to structured Koopman matrices and disentanglement. Overall, we propose a simple and easy to code new deep model that is fully unsupervised and it supports multifactor disentanglement. We showcase new disentangling abilities such as swapping of individual static factors between characters, and an incremental swap of disentangled factors from the source to the target. Moreover, we evaluate our method extensively on two factor standard benchmark tasks where we significantly improve over competing unsupervised approaches, and we perform competitively in comparison to weakly- and self-supervised state-of-the-art approaches. The code is available at https://github.com/azencot-group/SKD.
2023-03-30
{'id': '2303.17264v1', 'title': 'Multifactor Sequential Disentanglement via Structured Koopman Autoencoders', 'abstract': 'Disentangling complex data to its latent factors of variation is a\nfundamental task in representation learning. Existing work on sequential\ndisentanglement mostly provides two factor representations, i.e., it separates\nthe data to time-varying and time-invariant factors. In contrast, we consider\nmultifactor disentanglement in which multiple (more than two) semantic\ndisentangled components are generated. Key to our approach is a strong\ninductive bias where we assume that the underlying dynamics can be represented\nlinearly in the latent space. Under this assumption, it becomes natural to\nexploit the recently introduced Koopman autoencoder models. However,\ndisentangled representations are not guaranteed in Koopman approaches, and thus\nwe propose a novel spectral loss term which leads to structured Koopman\nmatrices and disentanglement. Overall, we propose a simple and easy to code new\ndeep model that is fully unsupervised and it supports multifactor\ndisentanglement. We showcase new disentangling abilities such as swapping of\nindividual static factors between characters, and an incremental swap of\ndisentangled factors from the source to the target. Moreover, we evaluate our\nmethod extensively on two factor standard benchmark tasks where we\nsignificantly improve over competing unsupervised approaches, and we perform\ncompetitively in comparison to weakly- and self-supervised state-of-the-art\napproaches. The code is available at https://github.com/azencot-group/SKD.', 'authors': ['Nimrod Berman', 'Ilan Naiman', 'Omri Azencot'], 'published': '2023-03-30 10:01:58+00:00', 'categories': ['cs.LG'], 'url': 'http://arxiv.org/abs/2303.17264v1'}
42,810
2305.13656v1
2,305.13656
1
Link Prediction without Graph Neural Networks
Link prediction, which consists of predicting edges based on graph features, is a fundamental task in many graph applications. As for several related problems, Graph Neural Networks (GNNs), which are based on an attribute-centric message-passing paradigm, have become the predominant framework for link prediction. GNNs have consistently outperformed traditional topology-based heuristics, but what contributes to their performance? Are there simpler approaches that achieve comparable or better results? To answer these questions, we first identify important limitations in how GNN-based link prediction methods handle the intrinsic class imbalance of the problem -- due to the graph sparsity -- in their training and evaluation. Moreover, we propose Gelato, a novel topology-centric framework that applies a topological heuristic to a graph enhanced by attribute information via graph learning. Our model is trained end-to-end with an N-pair loss on an unbiased training set to address class imbalance. Experiments show that Gelato is 145% more accurate, trains 11 times faster, infers 6,000 times faster, and has less than half of the trainable parameters compared to state-of-the-art GNNs for link prediction.
2023-05-23
{'id': '2305.13656v1', 'title': 'Link Prediction without Graph Neural Networks', 'abstract': 'Link prediction, which consists of predicting edges based on graph features,\nis a fundamental task in many graph applications. As for several related\nproblems, Graph Neural Networks (GNNs), which are based on an attribute-centric\nmessage-passing paradigm, have become the predominant framework for link\nprediction. GNNs have consistently outperformed traditional topology-based\nheuristics, but what contributes to their performance? Are there simpler\napproaches that achieve comparable or better results? To answer these\nquestions, we first identify important limitations in how GNN-based link\nprediction methods handle the intrinsic class imbalance of the problem -- due\nto the graph sparsity -- in their training and evaluation. Moreover, we propose\nGelato, a novel topology-centric framework that applies a topological heuristic\nto a graph enhanced by attribute information via graph learning. Our model is\ntrained end-to-end with an N-pair loss on an unbiased training set to address\nclass imbalance. Experiments show that Gelato is 145% more accurate, trains 11\ntimes faster, infers 6,000 times faster, and has less than half of the\ntrainable parameters compared to state-of-the-art GNNs for link prediction.', 'authors': ['Zexi Huang', 'Mert Kosan', 'Arlei Silva', 'Ambuj Singh'], 'published': '2023-05-23 03:59:21+00:00', 'categories': ['cs.LG', 'cs.SI'], 'url': 'http://arxiv.org/abs/2305.13656v1'}
42,811
2211.09066v1
2,211.09066
1
Teaching Algorithmic Reasoning via In-context Learning
Large language models (LLMs) have shown increasing in-context learning capabilities through scaling up model and data size. Despite this progress, LLMs are still unable to solve algorithmic reasoning problems. While providing a rationale with the final answer has led to further improvements in multi-step reasoning problems, Anil et al. 2022 showed that even simple algorithmic reasoning tasks such as parity are far from solved. In this work, we identify and study four key stages for successfully teaching algorithmic reasoning to LLMs: (1) formulating algorithms as skills, (2) teaching multiple skills simultaneously (skill accumulation), (3) teaching how to combine skills (skill composition) and (4) teaching how to use skills as tools. We show that it is possible to teach algorithmic reasoning to LLMs via in-context learning, which we refer to as algorithmic prompting. We evaluate our approach on a variety of arithmetic and quantitative reasoning tasks, and demonstrate significant boosts in performance over existing prompting techniques. In particular, for long parity, addition, multiplication and subtraction, we achieve an error reduction of approximately 10x, 9x, 5x and 2x respectively compared to the best available baselines.
2022-11-15
{'id': '2211.09066v1', 'title': 'Teaching Algorithmic Reasoning via In-context Learning', 'abstract': 'Large language models (LLMs) have shown increasing in-context learning\ncapabilities through scaling up model and data size. Despite this progress,\nLLMs are still unable to solve algorithmic reasoning problems. While providing\na rationale with the final answer has led to further improvements in multi-step\nreasoning problems, Anil et al. 2022 showed that even simple algorithmic\nreasoning tasks such as parity are far from solved. In this work, we identify\nand study four key stages for successfully teaching algorithmic reasoning to\nLLMs: (1) formulating algorithms as skills, (2) teaching multiple skills\nsimultaneously (skill accumulation), (3) teaching how to combine skills (skill\ncomposition) and (4) teaching how to use skills as tools. We show that it is\npossible to teach algorithmic reasoning to LLMs via in-context learning, which\nwe refer to as algorithmic prompting. We evaluate our approach on a variety of\narithmetic and quantitative reasoning tasks, and demonstrate significant boosts\nin performance over existing prompting techniques. In particular, for long\nparity, addition, multiplication and subtraction, we achieve an error reduction\nof approximately 10x, 9x, 5x and 2x respectively compared to the best available\nbaselines.', 'authors': ['Hattie Zhou', 'Azade Nova', 'Hugo Larochelle', 'Aaron Courville', 'Behnam Neyshabur', 'Hanie Sedghi'], 'published': '2022-11-15 06:12:28+00:00', 'categories': ['cs.LG', 'cs.AI', 'cs.CL'], 'url': 'http://arxiv.org/abs/2211.09066v1'}
42,812
2207.10751v2
2,207.10751
2
Federated Learning on Adaptively Weighted Nodes by Bilevel Optimization
We propose a federated learning method with weighted nodes in which the weights can be modified to optimize the model's performance on a separate validation set. The problem is formulated as a bilevel optimization where the inner problem is a federated learning problem with weighted nodes and the outer problem focuses on optimizing the weights based on the validation performance of the model returned from the inner problem. A communication-efficient federated optimization algorithm is designed to solve this bilevel optimization problem. Under an error-bound assumption, we analyze the generalization performance of the output model and identify scenarios when our method is in theory superior to training a model only locally and to federated learning with static and evenly distributed weights.
2022-07-21
{'id': '2207.10751v2', 'title': 'Federated Learning on Adaptively Weighted Nodes by Bilevel Optimization', 'abstract': "We propose a federated learning method with weighted nodes in which the\nweights can be modified to optimize the model's performance on a separate\nvalidation set. The problem is formulated as a bilevel optimization where the\ninner problem is a federated learning problem with weighted nodes and the outer\nproblem focuses on optimizing the weights based on the validation performance\nof the model returned from the inner problem. A communication-efficient\nfederated optimization algorithm is designed to solve this bilevel optimization\nproblem. Under an error-bound assumption, we analyze the generalization\nperformance of the output model and identify scenarios when our method is in\ntheory superior to training a model only locally and to federated learning with\nstatic and evenly distributed weights.", 'authors': ['Yankun Huang', 'Qihang Lin', 'Nick Street', 'Stephen Baek'], 'published': '2022-07-21 20:58:19+00:00', 'categories': ['cs.LG'], 'url': 'http://arxiv.org/abs/2207.10751v2'}
42,813
2111.05478v2
2,111.05478
2
SGD Through the Lens of Kolmogorov Complexity
We prove that stochastic gradient descent (SGD) finds a solution that achieves $(1-)$ classification accuracy on the entire dataset. We do so under two main assumptions: (1. Local progress) The model accuracy improves on average over batches. (2. Models compute simple functions) The function computed by the model is simple (has low Kolmogorov complexity). It is sufficient that these assumptions hold only for a tiny fraction of the epochs. Intuitively, the above means that intermittent local progress of SGD implies global progress. Assumption 2 trivially holds for underparameterized models, hence, our work gives the first convergence guarantee for general, underparameterized models. Furthermore, this is the first result which is completely model agnostic - we do not require the model to have any specific architecture or activation function, it may not even be a neural network. Our analysis makes use of the entropy compression method, which was first introduced by Moser and Tardos in the context of the Lov\'asz local lemma.
2021-11-10
{'id': '2111.05478v2', 'title': 'SGD Through the Lens of Kolmogorov Complexity', 'abstract': "We prove that stochastic gradient descent (SGD) finds a solution that\nachieves $(1-\\epsilon)$ classification accuracy on the entire dataset. We do so\nunder two main assumptions: (1. Local progress) The model accuracy improves on\naverage over batches. (2. Models compute simple functions) The function\ncomputed by the model is simple (has low Kolmogorov complexity). It is\nsufficient that these assumptions hold only for a tiny fraction of the epochs.\n Intuitively, the above means that intermittent local progress of SGD implies\nglobal progress. Assumption 2 trivially holds for underparameterized models,\nhence, our work gives the first convergence guarantee for general,\nunderparameterized models. Furthermore, this is the first result which is\ncompletely model agnostic - we do not require the model to have any specific\narchitecture or activation function, it may not even be a neural network. Our\nanalysis makes use of the entropy compression method, which was first\nintroduced by Moser and Tardos in the context of the Lov\\'asz local lemma.", 'authors': ['Gregory Schwartzman'], 'published': '2021-11-10 01:32:38+00:00', 'categories': ['cs.LG'], 'url': 'http://arxiv.org/abs/2111.05478v2'}
42,814
2206.15407v2
2,206.15407
2
Shifts 2.0: Extending The Dataset of Real Distributional Shifts
Distributional shift, or the mismatch between training and deployment data, is a significant obstacle to the usage of machine learning in high-stakes industrial applications, such as autonomous driving and medicine. This creates a need to be able to assess how robustly ML models generalize as well as the quality of their uncertainty estimates. Standard ML baseline datasets do not allow these properties to be assessed, as the training, validation and test data are often identically distributed. Recently, a range of dedicated benchmarks have appeared, featuring both distributionally matched and shifted data. Among these benchmarks, the Shifts dataset stands out in terms of the diversity of tasks as well as the data modalities it features. While most of the benchmarks are heavily dominated by 2D image classification tasks, Shifts contains tabular weather forecasting, machine translation, and vehicle motion prediction tasks. This enables the robustness properties of models to be assessed on a diverse set of industrial-scale tasks and either universal or directly applicable task-specific conclusions to be reached. In this paper, we extend the Shifts Dataset with two datasets sourced from industrial, high-risk applications of high societal importance. Specifically, we consider the tasks of segmentation of white matter Multiple Sclerosis lesions in 3D magnetic resonance brain images and the estimation of power consumption in marine cargo vessels. Both tasks feature ubiquitous distributional shifts and a strict safety requirement due to the high cost of errors. These new datasets will allow researchers to further explore robust generalization and uncertainty estimation in new situations. In this work, we provide a description of the dataset and baseline results for both tasks.
2022-06-30
{'id': '2206.15407v2', 'title': 'Shifts 2.0: Extending The Dataset of Real Distributional Shifts', 'abstract': 'Distributional shift, or the mismatch between training and deployment data,\nis a significant obstacle to the usage of machine learning in high-stakes\nindustrial applications, such as autonomous driving and medicine. This creates\na need to be able to assess how robustly ML models generalize as well as the\nquality of their uncertainty estimates. Standard ML baseline datasets do not\nallow these properties to be assessed, as the training, validation and test\ndata are often identically distributed. Recently, a range of dedicated\nbenchmarks have appeared, featuring both distributionally matched and shifted\ndata. Among these benchmarks, the Shifts dataset stands out in terms of the\ndiversity of tasks as well as the data modalities it features. While most of\nthe benchmarks are heavily dominated by 2D image classification tasks, Shifts\ncontains tabular weather forecasting, machine translation, and vehicle motion\nprediction tasks. This enables the robustness properties of models to be\nassessed on a diverse set of industrial-scale tasks and either universal or\ndirectly applicable task-specific conclusions to be reached. In this paper, we\nextend the Shifts Dataset with two datasets sourced from industrial, high-risk\napplications of high societal importance. Specifically, we consider the tasks\nof segmentation of white matter Multiple Sclerosis lesions in 3D magnetic\nresonance brain images and the estimation of power consumption in marine cargo\nvessels. Both tasks feature ubiquitous distributional shifts and a strict\nsafety requirement due to the high cost of errors. These new datasets will\nallow researchers to further explore robust generalization and uncertainty\nestimation in new situations. In this work, we provide a description of the\ndataset and baseline results for both tasks.', 'authors': ['Andrey Malinin', 'Andreas Athanasopoulos', 'Muhamed Barakovic', 'Meritxell Bach Cuadra', 'Mark J. F. Gales', 'Cristina Granziera', 'Mara Graziani', 'Nikolay Kartashev', 'Konstantinos Kyriakopoulos', 'Po-Jui Lu', 'Nataliia Molchanova', 'Antonis Nikitakis', 'Vatsal Raina', 'Francesco La Rosa', 'Eli Sivena', 'Vasileios Tsarsitalidis', 'Efi Tsompopoulou', 'Elena Volf'], 'published': '2022-06-30 16:51:52+00:00', 'categories': ['cs.LG', 'cs.AI', 'stat.ML'], 'url': 'http://arxiv.org/abs/2206.15407v2'}
42,815
2307.11228v1
2,307.11228
1
From Adaptive Query Release to Machine Unlearning
We formalize the problem of machine unlearning as design of efficient unlearning algorithms corresponding to learning algorithms which perform a selection of adaptive queries from structured query classes. We give efficient unlearning algorithms for linear and prefix-sum query classes. As applications, we show that unlearning in many problems, in particular, stochastic convex optimization (SCO), can be reduced to the above, yielding improved guarantees for the problem. In particular, for smooth Lipschitz losses and any $>0$, our results yield an unlearning algorithm with excess population risk of $O(1n+dn)$ with unlearning query (gradient) complexity $O(Retraining Complexity)$, where $d$ is the model dimensionality and $n$ is the initial number of samples. For non-smooth Lipschitz losses, we give an unlearning algorithm with excess population risk $O(1n+(dn)^1/2)$ with the same unlearning query (gradient) complexity. Furthermore, in the special case of Generalized Linear Models (GLMs), such as those in linear and logistic regression, we get dimension-independent rates of $O(1n +1(n)^2/3)$ and $O(1n +1(n)^1/3)$ for smooth Lipschitz and non-smooth Lipschitz losses respectively. Finally, we give generalizations of the above from one unlearning request to dynamic streams consisting of insertions and deletions.
2023-07-20
{'id': '2307.11228v1', 'title': 'From Adaptive Query Release to Machine Unlearning', 'abstract': 'We formalize the problem of machine unlearning as design of efficient\nunlearning algorithms corresponding to learning algorithms which perform a\nselection of adaptive queries from structured query classes. We give efficient\nunlearning algorithms for linear and prefix-sum query classes. As applications,\nwe show that unlearning in many problems, in particular, stochastic convex\noptimization (SCO), can be reduced to the above, yielding improved guarantees\nfor the problem. In particular, for smooth Lipschitz losses and any $\\rho>0$,\nour results yield an unlearning algorithm with excess population risk of\n$\\tilde O\\big(\\frac{1}{\\sqrt{n}}+\\frac{\\sqrt{d}}{n\\rho}\\big)$ with unlearning\nquery (gradient) complexity $\\tilde O(\\rho \\cdot \\text{Retraining\nComplexity})$, where $d$ is the model dimensionality and $n$ is the initial\nnumber of samples. For non-smooth Lipschitz losses, we give an unlearning\nalgorithm with excess population risk $\\tilde\nO\\big(\\frac{1}{\\sqrt{n}}+\\big(\\frac{\\sqrt{d}}{n\\rho}\\big)^{1/2}\\big)$ with the\nsame unlearning query (gradient) complexity. Furthermore, in the special case\nof Generalized Linear Models (GLMs), such as those in linear and logistic\nregression, we get dimension-independent rates of $\\tilde\nO\\big(\\frac{1}{\\sqrt{n}} +\\frac{1}{(n\\rho)^{2/3}}\\big)$ and $\\tilde\nO\\big(\\frac{1}{\\sqrt{n}} +\\frac{1}{(n\\rho)^{1/3}}\\big)$ for smooth Lipschitz\nand non-smooth Lipschitz losses respectively. Finally, we give generalizations\nof the above from one unlearning request to \\textit{dynamic} streams consisting\nof insertions and deletions.', 'authors': ['Enayat Ullah', 'Raman Arora'], 'published': '2023-07-20 20:46:39+00:00', 'categories': ['cs.LG', 'stat.ML'], 'url': 'http://arxiv.org/abs/2307.11228v1'}
42,816
2302.11831v1
2,302.11831
1
Embedding Fourier for Ultra-High-Definition Low-Light Image Enhancement
Ultra-High-Definition (UHD) photo has gradually become the standard configuration in advanced imaging devices. The new standard unveils many issues in existing approaches for low-light image enhancement (LLIE), especially in dealing with the intricate issue of joint luminance enhancement and noise removal while remaining efficient. Unlike existing methods that address the problem in the spatial domain, we propose a new solution, UHDFour, that embeds Fourier transform into a cascaded network. Our approach is motivated by a few unique characteristics in the Fourier domain: 1) most luminance information concentrates on amplitudes while noise is closely related to phases, and 2) a high-resolution image and its low-resolution version share similar amplitude patterns.Through embedding Fourier into our network, the amplitude and phase of a low-light image are separately processed to avoid amplifying noise when enhancing luminance. Besides, UHDFour is scalable to UHD images by implementing amplitude and phase enhancement under the low-resolution regime and then adjusting the high-resolution scale with few computations. We also contribute the first real UHD LLIE dataset, UHD-LL, that contains 2,150 low-noise/normal-clear 4K image pairs with diverse darkness and noise levels captured in different scenarios. With this dataset, we systematically analyze the performance of existing LLIE methods for processing UHD images and demonstrate the advantage of our solution. We believe our new framework, coupled with the dataset, would push the frontier of LLIE towards UHD. The code and dataset are available at https://li-chongyi.github.io/UHDFour.
2023-02-23
{'id': '2302.11831v1', 'title': 'Embedding Fourier for Ultra-High-Definition Low-Light Image Enhancement', 'abstract': 'Ultra-High-Definition (UHD) photo has gradually become the standard\nconfiguration in advanced imaging devices. The new standard unveils many issues\nin existing approaches for low-light image enhancement (LLIE), especially in\ndealing with the intricate issue of joint luminance enhancement and noise\nremoval while remaining efficient. Unlike existing methods that address the\nproblem in the spatial domain, we propose a new solution, UHDFour, that embeds\nFourier transform into a cascaded network. Our approach is motivated by a few\nunique characteristics in the Fourier domain: 1) most luminance information\nconcentrates on amplitudes while noise is closely related to phases, and 2) a\nhigh-resolution image and its low-resolution version share similar amplitude\npatterns.Through embedding Fourier into our network, the amplitude and phase of\na low-light image are separately processed to avoid amplifying noise when\nenhancing luminance. Besides, UHDFour is scalable to UHD images by implementing\namplitude and phase enhancement under the low-resolution regime and then\nadjusting the high-resolution scale with few computations. We also contribute\nthe first real UHD LLIE dataset, \\textbf{UHD-LL}, that contains 2,150\nlow-noise/normal-clear 4K image pairs with diverse darkness and noise levels\ncaptured in different scenarios. With this dataset, we systematically analyze\nthe performance of existing LLIE methods for processing UHD images and\ndemonstrate the advantage of our solution. We believe our new framework,\ncoupled with the dataset, would push the frontier of LLIE towards UHD. The code\nand dataset are available at https://li-chongyi.github.io/UHDFour.', 'authors': ['Chongyi Li', 'Chun-Le Guo', 'Man Zhou', 'Zhexin Liang', 'Shangchen Zhou', 'Ruicheng Feng', 'Chen Change Loy'], 'published': '2023-02-23 07:43:41+00:00', 'categories': ['cs.CV'], 'url': 'http://arxiv.org/abs/2302.11831v1'}
42,817
2206.13947v3
2,206.13947
3
Long Range Language Modeling via Gated State Spaces
State space models have shown to be effective at modeling long range dependencies, specially on sequence classification tasks. In this work we focus on autoregressive sequence modeling over English books, Github source code and ArXiv mathematics articles. Based on recent developments around the effectiveness of gated activation functions, we propose a new layer named Gated State Space (GSS) and show that it trains significantly faster than the diagonal version of S4 (i.e. DSS) on TPUs, is fairly competitive with several well-tuned Transformer-based baselines and exhibits zero-shot generalization to longer inputs while being straightforward to implement. Finally, we show that leveraging self-attention to model local dependencies improves the performance of GSS even further.
2022-06-27
{'id': '2206.13947v3', 'title': 'Long Range Language Modeling via Gated State Spaces', 'abstract': 'State space models have shown to be effective at modeling long range\ndependencies, specially on sequence classification tasks. In this work we focus\non autoregressive sequence modeling over English books, Github source code and\nArXiv mathematics articles. Based on recent developments around the\neffectiveness of gated activation functions, we propose a new layer named Gated\nState Space (GSS) and show that it trains significantly faster than the\ndiagonal version of S4 (i.e. DSS) on TPUs, is fairly competitive with several\nwell-tuned Transformer-based baselines and exhibits zero-shot generalization to\nlonger inputs while being straightforward to implement. Finally, we show that\nleveraging self-attention to model local dependencies improves the performance\nof GSS even further.', 'authors': ['Harsh Mehta', 'Ankit Gupta', 'Ashok Cutkosky', 'Behnam Neyshabur'], 'published': '2022-06-27 01:50:18+00:00', 'categories': ['cs.LG', 'cs.CL'], 'url': 'http://arxiv.org/abs/2206.13947v3'}
42,818
2301.11500v1
2,301.115
1
Understanding Incremental Learning of Gradient Descent: A Fine-grained Analysis of Matrix Sensing
It is believed that Gradient Descent (GD) induces an implicit bias towards good generalization in training machine learning models. This paper provides a fine-grained analysis of the dynamics of GD for the matrix sensing problem, whose goal is to recover a low-rank ground-truth matrix from near-isotropic linear measurements. It is shown that GD with small initialization behaves similarly to the greedy low-rank learning heuristics (Li et al., 2020) and follows an incremental learning procedure (Gissin et al., 2019): GD sequentially learns solutions with increasing ranks until it recovers the ground truth matrix. Compared to existing works which only analyze the first learning phase for rank-1 solutions, our result provides characterizations for the whole learning process. Moreover, besides the over-parameterized regime that many prior works focused on, our analysis of the incremental learning procedure also applies to the under-parameterized regime. Finally, we conduct numerical experiments to confirm our theoretical findings.
2023-01-27
{'id': '2301.11500v1', 'title': 'Understanding Incremental Learning of Gradient Descent: A Fine-grained Analysis of Matrix Sensing', 'abstract': 'It is believed that Gradient Descent (GD) induces an implicit bias towards\ngood generalization in training machine learning models. This paper provides a\nfine-grained analysis of the dynamics of GD for the matrix sensing problem,\nwhose goal is to recover a low-rank ground-truth matrix from near-isotropic\nlinear measurements. It is shown that GD with small initialization behaves\nsimilarly to the greedy low-rank learning heuristics (Li et al., 2020) and\nfollows an incremental learning procedure (Gissin et al., 2019): GD\nsequentially learns solutions with increasing ranks until it recovers the\nground truth matrix. Compared to existing works which only analyze the first\nlearning phase for rank-1 solutions, our result provides characterizations for\nthe whole learning process. Moreover, besides the over-parameterized regime\nthat many prior works focused on, our analysis of the incremental learning\nprocedure also applies to the under-parameterized regime. Finally, we conduct\nnumerical experiments to confirm our theoretical findings.', 'authors': ['Jikai Jin', 'Zhiyuan Li', 'Kaifeng Lyu', 'Simon S. Du', 'Jason D. Lee'], 'published': '2023-01-27 02:30:51+00:00', 'categories': ['cs.LG', 'math.OC', 'stat.ML'], 'url': 'http://arxiv.org/abs/2301.11500v1'}
42,819
2210.03070v2
2,210.0307
2
Toxicity in Multilingual Machine Translation at Scale
Machine Translation systems can produce different types of errors, some of which are characterized as critical or catastrophic due to the specific negative impact that they can have on users. In this paper we focus on one type of critical error: added toxicity. We evaluate and analyze added toxicity when translating a large evaluation dataset (HOLISTICBIAS, over 472k sentences, covering 13 demographic axes) from English into 164 languages. An automatic toxicity evaluation shows that added toxicity across languages varies from 0% to 5%. The output languages with the most added toxicity tend to be low-resource ones, and the demographic axes with the most added toxicity include sexual orientation, gender and sex, and ability. We also perform human evaluation on a subset of 8 translation directions, confirming the prevalence of true added toxicity. We use a measurement of the amount of source contribution to the translation, where a low source contribution implies hallucination, to interpret what causes toxicity. Making use of the input attributions allows us to explain toxicity, because the source contributions significantly correlate with toxicity for 84% of languages studied. Given our findings, our recommendations to reduce added toxicity are to curate training data to avoid mistranslations, mitigate hallucination and check unstable translations.
2022-10-06
{'id': '2210.03070v2', 'title': 'Toxicity in Multilingual Machine Translation at Scale', 'abstract': 'Machine Translation systems can produce different types of errors, some of\nwhich are characterized as critical or catastrophic due to the specific\nnegative impact that they can have on users. In this paper we focus on one type\nof critical error: added toxicity. We evaluate and analyze added toxicity when\ntranslating a large evaluation dataset (HOLISTICBIAS, over 472k sentences,\ncovering 13 demographic axes) from English into 164 languages. An automatic\ntoxicity evaluation shows that added toxicity across languages varies from 0%\nto 5%. The output languages with the most added toxicity tend to be\nlow-resource ones, and the demographic axes with the most added toxicity\ninclude sexual orientation, gender and sex, and ability. We also perform human\nevaluation on a subset of 8 translation directions, confirming the prevalence\nof true added toxicity. We use a measurement of the amount of source\ncontribution to the translation, where a low source contribution implies\nhallucination, to interpret what causes toxicity. Making use of the input\nattributions allows us to explain toxicity, because the source contributions\nsignificantly correlate with toxicity for 84% of languages studied. Given our\nfindings, our recommendations to reduce added toxicity are to curate training\ndata to avoid mistranslations, mitigate hallucination and check unstable\ntranslations.', 'authors': ['Marta R. Costa-jussà', 'Eric Smith', 'Christophe Ropers', 'Daniel Licht', 'Jean Maillard', 'Javier Ferrando', 'Carlos Escolano'], 'published': '2022-10-06 17:26:27+00:00', 'categories': ['cs.CL', 'I.2.7'], 'url': 'http://arxiv.org/abs/2210.03070v2'}
42,820
2305.04073v2
2,305.04073
2
Explaining RL Decisions with Trajectories
Explanation is a key component for the adoption of reinforcement learning (RL) in many real-world decision-making problems. In the literature, the explanation is often provided by saliency attribution to the features of the RL agent's state. In this work, we propose a complementary approach to these explanations, particularly for offline RL, where we attribute the policy decisions of a trained RL agent to the trajectories encountered by it during training. To do so, we encode trajectories in offline training data individually as well as collectively (encoding a set of trajectories). We then attribute policy decisions to a set of trajectories in this encoded space by estimating the sensitivity of the decision with respect to that set. Further, we demonstrate the effectiveness of the proposed approach in terms of quality of attributions as well as practical scalability in diverse environments that involve both discrete and continuous state and action spaces such as grid-worlds, video games (Atari) and continuous control (MuJoCo). We also conduct a human study on a simple navigation task to observe how their understanding of the task compares with data attributed for a trained RL policy. Keywords -- Explainable AI, Verifiability of AI Decisions, Explainable RL.
2023-05-06
{'id': '2305.04073v2', 'title': 'Explaining RL Decisions with Trajectories', 'abstract': "Explanation is a key component for the adoption of reinforcement learning\n(RL) in many real-world decision-making problems. In the literature, the\nexplanation is often provided by saliency attribution to the features of the RL\nagent's state. In this work, we propose a complementary approach to these\nexplanations, particularly for offline RL, where we attribute the policy\ndecisions of a trained RL agent to the trajectories encountered by it during\ntraining. To do so, we encode trajectories in offline training data\nindividually as well as collectively (encoding a set of trajectories). We then\nattribute policy decisions to a set of trajectories in this encoded space by\nestimating the sensitivity of the decision with respect to that set. Further,\nwe demonstrate the effectiveness of the proposed approach in terms of quality\nof attributions as well as practical scalability in diverse environments that\ninvolve both discrete and continuous state and action spaces such as\ngrid-worlds, video games (Atari) and continuous control (MuJoCo). We also\nconduct a human study on a simple navigation task to observe how their\nunderstanding of the task compares with data attributed for a trained RL\npolicy. Keywords -- Explainable AI, Verifiability of AI Decisions, Explainable\nRL.", 'authors': ['Shripad Vilasrao Deshmukh', 'Arpan Dasgupta', 'Balaji Krishnamurthy', 'Nan Jiang', 'Chirag Agarwal', 'Georgios Theocharous', 'Jayakumar Subramanian'], 'published': '2023-05-06 15:26:22+00:00', 'categories': ['cs.AI', 'cs.LG'], 'url': 'http://arxiv.org/abs/2305.04073v2'}
42,821
2210.05359v1
2,210.05359
1
Mind's Eye: Grounded Language Model Reasoning through Simulation
Successful and effective communication between humans and AI relies on a shared experience of the world. By training solely on written text, current language models (LMs) miss the grounded experience of humans in the real-world -- their failure to relate language to the physical world causes knowledge to be misrepresented and obvious mistakes in their reasoning. We present Mind's Eye, a paradigm to ground language model reasoning in the physical world. Given a physical reasoning question, we use a computational physics engine (DeepMind's MuJoCo) to simulate the possible outcomes, and then use the simulation results as part of the input, which enables language models to perform reasoning. Experiments on 39 tasks in a physics alignment benchmark demonstrate that Mind's Eye can improve reasoning ability by a large margin (27.9% zero-shot, and 46.0% few-shot absolute accuracy improvement on average). Smaller language models armed with Mind's Eye can obtain similar performance to models that are 100x larger. Finally, we confirm the robustness of Mind's Eye through ablation studies.
2022-10-11
{'id': '2210.05359v1', 'title': "Mind's Eye: Grounded Language Model Reasoning through Simulation", 'abstract': "Successful and effective communication between humans and AI relies on a\nshared experience of the world. By training solely on written text, current\nlanguage models (LMs) miss the grounded experience of humans in the real-world\n-- their failure to relate language to the physical world causes knowledge to\nbe misrepresented and obvious mistakes in their reasoning. We present Mind's\nEye, a paradigm to ground language model reasoning in the physical world. Given\na physical reasoning question, we use a computational physics engine\n(DeepMind's MuJoCo) to simulate the possible outcomes, and then use the\nsimulation results as part of the input, which enables language models to\nperform reasoning. Experiments on 39 tasks in a physics alignment benchmark\ndemonstrate that Mind's Eye can improve reasoning ability by a large margin\n(27.9% zero-shot, and 46.0% few-shot absolute accuracy improvement on average).\nSmaller language models armed with Mind's Eye can obtain similar performance to\nmodels that are 100x larger. Finally, we confirm the robustness of Mind's Eye\nthrough ablation studies.", 'authors': ['Ruibo Liu', 'Jason Wei', 'Shixiang Shane Gu', 'Te-Yen Wu', 'Soroush Vosoughi', 'Claire Cui', 'Denny Zhou', 'Andrew M. Dai'], 'published': '2022-10-11 11:39:23+00:00', 'categories': ['cs.CL', 'cs.AI'], 'url': 'http://arxiv.org/abs/2210.05359v1'}
42,822
2301.01609v2
2,301.01609
2
Emergent collective intelligence from massive-agent cooperation and competition
Inspired by organisms evolving through cooperation and competition between different populations on Earth, we study the emergence of artificial collective intelligence through massive-agent reinforcement learning. To this end, We propose a new massive-agent reinforcement learning environment, Lux, where dynamic and massive agents in two teams scramble for limited resources and fight off the darkness. In Lux, we build our agents through the standard reinforcement learning algorithm in curriculum learning phases and leverage centralized control via a pixel-to-pixel policy network. As agents co-evolve through self-play, we observe several stages of intelligence, from the acquisition of atomic skills to the development of group strategies. Since these learned group strategies arise from individual decisions without an explicit coordination mechanism, we claim that artificial collective intelligence emerges from massive-agent cooperation and competition. We further analyze the emergence of various learned strategies through metrics and ablation studies, aiming to provide insights for reinforcement learning implementations in massive-agent environments.
2023-01-04
{'id': '2301.01609v2', 'title': 'Emergent collective intelligence from massive-agent cooperation and competition', 'abstract': 'Inspired by organisms evolving through cooperation and competition between\ndifferent populations on Earth, we study the emergence of artificial collective\nintelligence through massive-agent reinforcement learning. To this end, We\npropose a new massive-agent reinforcement learning environment, Lux, where\ndynamic and massive agents in two teams scramble for limited resources and\nfight off the darkness. In Lux, we build our agents through the standard\nreinforcement learning algorithm in curriculum learning phases and leverage\ncentralized control via a pixel-to-pixel policy network. As agents co-evolve\nthrough self-play, we observe several stages of intelligence, from the\nacquisition of atomic skills to the development of group strategies. Since\nthese learned group strategies arise from individual decisions without an\nexplicit coordination mechanism, we claim that artificial collective\nintelligence emerges from massive-agent cooperation and competition. We further\nanalyze the emergence of various learned strategies through metrics and\nablation studies, aiming to provide insights for reinforcement learning\nimplementations in massive-agent environments.', 'authors': ['Hanmo Chen', 'Stone Tao', 'Jiaxin Chen', 'Weihan Shen', 'Xihui Li', 'Chenghui Yu', 'Sikai Cheng', 'Xiaolong Zhu', 'Xiu Li'], 'published': '2023-01-04 13:23:12+00:00', 'categories': ['cs.AI', 'cs.MA'], 'url': 'http://arxiv.org/abs/2301.01609v2'}
42,823
2302.07253v2
2,302.07253
2
Energy Transformer
Our work combines aspects of three promising paradigms in machine learning, namely, attention mechanism, energy-based models, and associative memory. Attention is the power-house driving modern deep learning successes, but it lacks clear theoretical foundations. Energy-based models allow a principled approach to discriminative and generative tasks, but the design of the energy functional is not straightforward. At the same time, Dense Associative Memory models or Modern Hopfield Networks have a well-established theoretical foundation, and allow an intuitive design of the energy function. We propose a novel architecture, called the Energy Transformer (or ET for short), that uses a sequence of attention layers that are purposely designed to minimize a specifically engineered energy function, which is responsible for representing the relationships between the tokens. In this work, we introduce the theoretical foundations of ET, explore its empirical capabilities using the image completion task, and obtain strong quantitative results on the graph anomaly detection and graph classification tasks.
2023-02-14
{'id': '2302.07253v2', 'title': 'Energy Transformer', 'abstract': 'Our work combines aspects of three promising paradigms in machine learning,\nnamely, attention mechanism, energy-based models, and associative memory.\nAttention is the power-house driving modern deep learning successes, but it\nlacks clear theoretical foundations. Energy-based models allow a principled\napproach to discriminative and generative tasks, but the design of the energy\nfunctional is not straightforward. At the same time, Dense Associative Memory\nmodels or Modern Hopfield Networks have a well-established theoretical\nfoundation, and allow an intuitive design of the energy function. We propose a\nnovel architecture, called the Energy Transformer (or ET for short), that uses\na sequence of attention layers that are purposely designed to minimize a\nspecifically engineered energy function, which is responsible for representing\nthe relationships between the tokens. In this work, we introduce the\ntheoretical foundations of ET, explore its empirical capabilities using the\nimage completion task, and obtain strong quantitative results on the graph\nanomaly detection and graph classification tasks.', 'authors': ['Benjamin Hoover', 'Yuchen Liang', 'Bao Pham', 'Rameswar Panda', 'Hendrik Strobelt', 'Duen Horng Chau', 'Mohammed J. Zaki', 'Dmitry Krotov'], 'published': '2023-02-14 18:51:22+00:00', 'categories': ['cs.LG', 'cond-mat.dis-nn', 'cs.CV', 'q-bio.NC', 'stat.ML'], 'url': 'http://arxiv.org/abs/2302.07253v2'}
42,824
2305.01034v2
2,305.01034
2
Model-agnostic Measure of Generalization Difficulty
The measure of a machine learning algorithm is the difficulty of the tasks it can perform, and sufficiently difficult tasks are critical drivers of strong machine learning models. However, quantifying the generalization difficulty of machine learning benchmarks has remained challenging. We propose what is to our knowledge the first model-agnostic measure of the inherent generalization difficulty of tasks. Our inductive bias complexity measure quantifies the total information required to generalize well on a task minus the information provided by the data. It does so by measuring the fractional volume occupied by hypotheses that generalize on a task given that they fit the training data. It scales exponentially with the intrinsic dimensionality of the space over which the model must generalize but only polynomially in resolution per dimension, showing that tasks which require generalizing over many dimensions are drastically more difficult than tasks involving more detail in fewer dimensions. Our measure can be applied to compute and compare supervised learning, reinforcement learning and meta-learning generalization difficulties against each other. We show that applied empirically, it formally quantifies intuitively expected trends, e.g. that in terms of required inductive bias, MNIST < CIFAR10 < Imagenet and fully observable Markov decision processes (MDPs) < partially observable MDPs. Further, we show that classification of complex images < few-shot meta-learning with simple images. Our measure provides a quantitative metric to guide the construction of more complex tasks requiring greater inductive bias, and thereby encourages the development of more sophisticated architectures and learning algorithms with more powerful generalization capabilities.
2023-05-01
{'id': '2305.01034v2', 'title': 'Model-agnostic Measure of Generalization Difficulty', 'abstract': 'The measure of a machine learning algorithm is the difficulty of the tasks it\ncan perform, and sufficiently difficult tasks are critical drivers of strong\nmachine learning models. However, quantifying the generalization difficulty of\nmachine learning benchmarks has remained challenging. We propose what is to our\nknowledge the first model-agnostic measure of the inherent generalization\ndifficulty of tasks. Our inductive bias complexity measure quantifies the total\ninformation required to generalize well on a task minus the information\nprovided by the data. It does so by measuring the fractional volume occupied by\nhypotheses that generalize on a task given that they fit the training data. It\nscales exponentially with the intrinsic dimensionality of the space over which\nthe model must generalize but only polynomially in resolution per dimension,\nshowing that tasks which require generalizing over many dimensions are\ndrastically more difficult than tasks involving more detail in fewer\ndimensions. Our measure can be applied to compute and compare supervised\nlearning, reinforcement learning and meta-learning generalization difficulties\nagainst each other. We show that applied empirically, it formally quantifies\nintuitively expected trends, e.g. that in terms of required inductive bias,\nMNIST < CIFAR10 < Imagenet and fully observable Markov decision processes\n(MDPs) < partially observable MDPs. Further, we show that classification of\ncomplex images < few-shot meta-learning with simple images. Our measure\nprovides a quantitative metric to guide the construction of more complex tasks\nrequiring greater inductive bias, and thereby encourages the development of\nmore sophisticated architectures and learning algorithms with more powerful\ngeneralization capabilities.', 'authors': ['Akhilan Boopathy', 'Kevin Liu', 'Jaedong Hwang', 'Shu Ge', 'Asaad Mohammedsaleh', 'Ila Fiete'], 'published': '2023-05-01 18:48:55+00:00', 'categories': ['cs.LG', 'cs.AI', 'stat.ML'], 'url': 'http://arxiv.org/abs/2305.01034v2'}