conference
stringclasses 6
values | title
stringlengths 2
176
| abstract
stringlengths 2
5k
| decision
stringclasses 11
values |
|---|---|---|---|
ICLR.cc/2023/Conference | SYNC: Efficient Neural Code Search Through Structurally Guided Hard Negative Curricula | Efficient code snippet search using natural language queries can be a great productivity tool for developers (beginners and professionals alike). Recently neural code search has been popular, where a neural method is used to embed both the query (NL) and the code snippet (PL) into a common representation space; which is further used to obtain the most relevant PL satisfying the intent in the query. Transformers-based pre-trained language models (such as CodeBERT, GraphCodeBERT, UniXCoder) have been especially effective to learn such representation. These models often make mistakes such as retrieving snippets with incorrect data types, and incorrect method names or signatures; even when exposed to the underlying structural information of the code (such as Abstract Syntax Tree and other static analysis outputs) during pre-training. The generalization ability beyond the training data is also limited (as the code retrieval datasets vary in the ways NL-PL pairs are collected). In this work, we propose a structure-aware hard negative sampling method and a mastering-rate based curriculum learning technique (SYNC) that enhances the pre-trained representation using both soft (random) and the (synthesized) hard negative samples. Our experiments on three state-of-the-art pre-trained language models for programming languages, over four Python code retrieval datasets, show the efficacy of the approach (under both in-distribution and out-of-distribution settings). | Withdrawn |
ICLR.cc/2022/Conference | Achieving Small-Batch Accuracy with Large-Batch Scalability via Adaptive Learning Rate Adjustment | We consider synchronous data-parallel neural network training with fixed large batch sizes. While the large batch size provides a high degree of parallelism, it likely degrades the generalization performance due to the low gradient noise scale. We propose a two-phase adaptive learning rate adjustment framework that tackles the poor generalization issue in large-batch training. Our empirical study shows that the number of training epochs before decaying the learning rate strongly affects the final accuracy. The framework performs extra epochs using the large learning rate even after the loss is flattened. After sufficient training under the noisy condition, the framework decays the learning rate based on the observed loss landscape at run-time. Our experimental results demonstrate that the proposed heuristics and algorithm enable to use an extremely large batch size while maintaining the model accuracy. For CIFAR-10 classification with ResNet20, our method achieves $92.66\%$ accuracy using $8,192$ batch size, which is close to $92.83\%$ achieved using $128$ batch size, at a negligible extra computational cost. | Withdrawn |
ICLR.cc/2021/Conference | Conditioning Trick for Training Stable GANs | In this paper we propose a conditioning trick, called difference departure from normality, applied on the generator network in response to instability issues during GAN training. We force the generator to get closer to the departure from normality function of real samples computed in the spectral domain of Schur decomposition. This binding makes the generator amenable to truncation and does not limit exploring all the possible modes. We slightly modify the BigGAN architecture incorporating residual network for synthesizing 2D representations of audio signals which enables reconstructing high quality sounds with some preserved phase information. Additionally, the proposed conditional training scenario makes a trade-off between fidelity and variety for the generated spectrograms. The experimental results on UrbanSound8k and ESC-50 environmental sound datasets and the Mozilla common voice dataset have shown that the proposed GAN configuration with the conditioning trick remarkably outperforms baseline architectures, according to three objective metrics: inception score, Frechet inception distance, and signal-to-noise ratio. | Withdrawn |
ICLR.cc/2023/Conference | CORE-PERIPHERY PRINCIPLE GUIDED REDESIGN OF SELF-ATTENTION IN TRANSFORMERS | Designing more efficient, reliable, and explainable neural network architectures is a crucial topic in the artificial intelligence (AI) field. Numerous efforts have been devoted to exploring the best structures, or structural signatures, of well-performing artificial neural networks (ANN). Previous studies, by post-hoc analysis, have found that the best-performing ANNs surprisingly resemble biological neural networks (BNN), which indicates that ANNs and BNNs may share some common principles to achieve optimal performance in either machine learning tasks or cognitive/behavior processes. Inspired by this phenomenon, rather than relying on post-hoc schemes, we proactively instill organizational principles of BNNs to guide the redesign of ANNs by infusing an efficient information communication mechanism of BNNs into ANNs. Specifically, we quantified the typical Core-Periphery (CP) organization of the human brain networks, infused the CorePeriphery principle into the redesign of the vision transformer (ViT), and proposed a novel CP-ViT architecture: the pair-wised densely interconnected self-attention architecture of ViT was upgraded by a sparse Core-Periphery architecture. In CPViT, the attention operation between nodes (image patches) is defined by a sparse graph with a Core-Periphery structure (CP graph), where the core nodes are redesigned and reorganized to play an integrative role and serve as a center for other periphery nodes to exchange information. We evaluated the proposed CP-ViT on multiple public datasets, including medical image datasets (INbreast) and natural image datasets (CIFAR-100). We show that there exist sweet spots of CP graphs that lead to CP-ViTs with significantly improved performance. In general, our work advances the state of the art in three aspects: 1) This work provides novel insights for brain-inspired AI: we can instill the efficient information communication mechanism of BNNs into ANNs by infusing similar organizational principles of BNNs into ANNs; 2) The optimized CP-ViT can significantly improve its predictive performance while dramatically reducing computational cost by benefiting from the infused efficient information communication mechanism existing in BNNs; and 3) The core nodes in CP-ViT can identify task-related meaningful and important image patches, which can significantly enhance the interpretability of the trained deep model. (Code is ready for release). | Withdrawn |
ICLR.cc/2019/Conference | Object detection deep learning networks for Optical Character Recognition | In this article, we show how we applied a simple approach coming from deep learning networks for object detection to the task of optical character recognition in order to build image features taylored for documents. In contrast to scene text reading in natural images using networks pretrained on ImageNet, our document reading is performed with small networks inspired by MNIST digit recognition challenge, at a small computational budget and a small stride. The object detection modern frameworks allow a direct end-to-end training, with no other algorithm than the deep learning and the non-max-suppression algorithm to filter the duplicate predictions. The trained weights can be used for higher level models, such as, for example, document classification, or document segmentation.
| Reject |
ICLR.cc/2022/Conference | StARformer: Transformer with State-Action-Reward Representations | Reinforcement Learning (RL) can be considered as a sequence modeling task, i.e., given a sequence of past state-action-reward experiences, a model autoregressively predicts a sequence of future actions. Recently, Transformers have been successfully adopted to model this problem. In this work, we propose State-Action-Reward Transformer (StARformer), which explicitly models strongly related local causal relations to help improve action prediction in long sequences. StARformer first extracts local representations (i.e., StAR-representations) from each group of state-action-reward tokens within a very short time span. A sequence of such local representations combined with state representations, is then used to make action predictions over a long time span. Our experiments show that StARformer outperforms the state-of-the-art Transformer-based method on Atari (image) and Gym (state vector) benchmarks, in both offline-RL and imitation learning settings. StARformer is also more compliant with longer sequences of inputs compared to the baseline. The code will be released online. | Withdrawn |
ICLR.cc/2021/Conference | PIVEN: A Deep Neural Network for Prediction Intervals with Specific Value Prediction | Improving the robustness of neural nets in regression tasks is key to their application in multiple domains. Deep learning-based approaches aim to achieve this goal either by improving their prediction of specific values (i.e., point prediction), or by producing prediction intervals (PIs) that quantify uncertainty. We present PIVEN, a deep neural network for producing both a PI and a prediction of specific values. Unlike previous studies, PIVEN makes no assumptions regarding data distribution inside the PI, making its point prediction more effective for various real-world problems. Benchmark experiments show that our approach produces tighter uncertainty bounds than the current state-of-the-art approach for producing PIs, while maintaining comparable performance to the state-of-the-art approach for specific value-prediction. Additional evaluation on large image datasets further support our conclusions. | Reject |
ICLR.cc/2022/Conference | On Hard Episodes in Meta-Learning | Existing meta-learners primarily focus on improving the average task accuracy across multiple episodes. Different episodes, however, may vary in hardness and quality leading to a wide gap in the meta-learner's performance across episodes. Understanding this issue is particularly critical in industrial few-shot settings, where there is limited control over test episodes as they are typically uploaded by end-users. In this paper, we empirically analyse the behaviour of meta-learners on episodes of varying hardness across three standard benchmark datasets: CIFAR-FS, mini-ImageNet, and tiered-ImageNet. Surprisingly, we observe a wide gap in accuracy of around $50\%$ between the hardest and easiest episodes across all the standard benchmarks and meta-learners. We additionally investigate various properties of hard episodes and highlight their connection to catastrophic forgetting during meta-training. To address the issue of sub-par performance on hard episodes, we investigate and benchmark different meta-training strategies based on adversarial training and curriculum learning. We find that adversarial training strategies are much more powerful than curriculum learning in improving the prediction performance on hard episodes. | Reject |
ICLR.cc/2021/Conference | Neural ODE Processes | Neural Ordinary Differential Equations (NODEs) use a neural network to model the instantaneous rate of change in the state of a system. However, despite their apparent suitability for dynamics-governed time-series, NODEs present a few disadvantages. First, they are unable to adapt to incoming data-points, a fundamental requirement for real-time applications imposed by the natural direction of time. Second, time-series are often composed of a sparse set of measurements that could be explained by many possible underlying dynamics. NODEs do not capture this uncertainty. In contrast, Neural Processes (NPs) are a new class of stochastic processes providing uncertainty estimation and fast data-adaptation, but lack an explicit treatment of the flow of time. To address these problems, we introduce Neural ODE Processes (NDPs), a new class of stochastic processes determined by a distribution over Neural ODEs. By maintaining an adaptive data-dependent distribution over the underlying ODE, we show that our model can successfully capture the dynamics of low-dimensional systems from just a few data-points. At the same time, we demonstrate that NDPs scale up to challenging high-dimensional time-series with unknown latent dynamics such as rotating MNIST digits. | Accept (Poster) |
ICLR.cc/2023/Conference | Unsupervised Threshold Learning with "$L$"-trend Prior For Visual Anomaly Detection | This paper considers unsupervised threshold learning, a practical yet under-researched module of anomaly detection (AD) for image data. AD comprises two separate modules: score generation and threshold learning. Most existing studies are more curious about the first part. It is often assumed that if the scoring module is good, estimating an accurate threshold is within easy reach. However, we argue that in the context of computer vision, some challenges in high-dimensional space lead threshold estimation be a non-trivial problem. In this paper, we leverage the inherent difference between normal instances and anomalies by ranking their anomaly score, which shows a phenomenon that involves two distinct trends. We term it as the "$L$"-trend prior. With that finding, we utilize an adaptive polynomial regression model to determine the threshold. Unlike the classic threshold learners which rely on enough training samples or statistical assumptions, this method is plug-and-play that can be implemented into different anomaly score function among various datasets. Also, the evaluation results demonstrate an obvious improvement. | Withdrawn |
ICLR.cc/2018/Conference | GraphVAE: Towards Generation of Small Graphs Using Variational Autoencoders | Deep learning on graphs has become a popular research topic with many applications. However, past work has concentrated on learning graph embedding tasks only, which is in contrast with advances in generative models for images and text. Is it possible to transfer this progress to the domain of graphs? We propose to sidestep hurdles associated with linearization of such discrete structures by having a decoder output a probabilistic fully-connected graph of a predefined maximum size directly at once. Our method is formulated as a variational autoencoder. We evaluate on the challenging task of conditional molecule generation. | Reject |
ICLR.cc/2019/Conference | Doubly Sparse: Sparse Mixture of Sparse Experts for Efficient Softmax Inference | Computations for the softmax function in neural network models are expensive when the number of output classes is large. This can become a significant issue in both training and inference for such models. In this paper, we present Doubly Sparse Softmax (DS-Softmax), Sparse Mixture of Sparse of Sparse Experts, to improve the efficiency for softmax inference. During training, our method learns a two-level class hierarchy by dividing entire output class space into several partially overlapping experts. Each expert is responsible for a learned subset of the output class space and each output class only belongs to a small number of those experts. During inference, our method quickly locates the most probable expert to compute small-scale softmax. Our method is learning-based and requires no knowledge of the output class partition space a priori. We empirically evaluate our method on several real-world tasks and demonstrate that we can achieve significant computation reductions without loss of performance. | Reject |
ICLR.cc/2023/Conference | When does Bias Transfer in Transfer Learning? | Using transfer learning to adapt a pre-trained "source model" to a downstream "target task" can dramatically increase performance with seemingly no downside. In this work, we demonstrate that there can exist a downside after all: bias transfer, or the tendency for biases of the source model to persist even after adapting the model to the target class. Through a combination of synthetic and natural experiments, we show that bias transfer both (a) arises in realistic settings (such as when pre-training on ImageNet or other standard datasets) and (b) can occur even when the target dataset is explicitly de-biased. As transfer-learned models are increasingly deployed in the real world, our work highlights the importance of understanding the limitations of pre-trained source models. | Withdrawn |
ICLR.cc/2023/Conference | Deep Leakage from Model in Federated Learning | Distributed machine learning has been widely used in recent years to tackle large and complex dataset problems. Therewith, the security of distributed learning has also drawn increasing attention from both academia and industry. In this context, federated learning (FL) was developed as a “secure” distributed learning by maintaining private training data locally and only public model gradients are communicated between. However, to date, a variety of gradient leakage attacks have been proposed for this procedure and prove that it is insecure. For instance, a common drawback of these attacks is shared: {they require} too much auxiliary information such as model weights, optimizers, and some hyperparameters (e.g., learning rate), which are difficult to obtain in real situations. Moreover, many existing algorithms avoid transmitting model gradients in FL and turn to sending model weights, such as FedAvg, but few people consider its security breach. In this paper, we present two novel frameworks to demonstrate that transmitting model weights is also likely to leak private local data of clients, i.e., (DLM and DLM+), under the FL scenario. In addition, a variety of experiments are performed to illustrate the effect and generality of our attack frameworks. At the end of this paper, we also introduce two defenses to the proposed attacks and evaluate their protective effects. Comprehensively, the proposed attack and defense schemes can be applied to the generally distributed learning scenario as well, just with some appropriate customization. | Withdrawn |
ICLR.cc/2018/Conference | Now I Remember! Episodic Memory For Reinforcement Learning | Humans rely on episodic memory constantly, in remembering the name of someone they met 10 minutes ago, the plot of a movie as it unfolds, or where they parked the car. Endowing reinforcement learning agents with episodic memory is a key step on the path toward replicating human-like general intelligence. We analyze why standard RL agents lack episodic memory today, and why existing RL tasks don't require it. We design a new form of external memory called Masked Experience Memory, or MEM, modeled after key features of human episodic memory. To evaluate episodic memory we define an RL task based on the common children's game of Concentration. We find that a MEM RL agent leverages episodic memory effectively to master Concentration, unlike the baseline agents we tested. | Reject |
ICLR.cc/2021/Conference | Probabilistic Multimodal Representation Learning | Learning multimodal representations is a requirement for many tasks such as image--caption retrieval. Previous work on this problem has only focused on finding good vector representations without any explicit measure of uncertainty. In this work, we argue and demonstrate that learning multimodal representations as probability distributions can lead to better representations, as well as providing other benefits such as adding a measure of uncertainty to the learned representations. We show that this measure of uncertainty can capture how confident our model is about the representations in the multimodal domain, i.e, how clear it is for the model to retrieve/predict the matching pair. We experiment with similarity metrics that have not been traditionally used for the multimodal retrieval task, and show that the choice of the similarity metric affects the quality of the learned representations. | Withdrawn |
ICLR.cc/2020/Conference | Factorized Multimodal Transformer for Multimodal Sequential Learning | The complex world around us is inherently multimodal and sequential (continuous). Information is scattered across different modalities and requires multiple continuous sensors to be captured. As machine learning leaps towards better generalization to real world, multimodal sequential learning becomes a fundamental research area. Arguably, modeling arbitrarily distributed spatio-temporal dynamics within and across modalities is the biggest challenge in this research area. In this paper, we present a new transformer model, called the Factorized Multimodal Transformer (FMT) for multimodal sequential learning. FMT inherently models the intramodal and intermodal (involving two or more modalities) dynamics within its multimodal input in a factorized manner. The proposed factorization allows for increasing the number of self-attentions to better model the multimodal phenomena at hand; without encountering difficulties during training (e.g. overfitting) even on relatively low-resource setups. All the attention mechanisms within FMT have a full time-domain receptive field which allows them to asynchronously capture long-range multimodal dynamics. In our experiments we focus on datasets that contain the three commonly studied modalities of language, vision and acoustic. We perform a wide range of experiments, spanning across 3 well-studied datasets and 21 distinct labels. FMT shows superior performance over previously proposed models, setting new state of the art in the studied datasets. | Withdrawn |
ICLR.cc/2021/Conference | Everybody's Talkin': Let Me Talk as You Want | We present a method to edit a target portrait footage by taking a sequence of audio as input to synthesize a photo-realistic video. This method is unique because it is highly dynamic. It does not assume a person-specific rendering network yet capable of translating one source audio into one random chosen video output within a set of speech videos. Instead of learning a highly heterogeneous and nonlinear mapping from audio to the video directly, we first factorize each target video frame into orthogonal parameter spaces, i.e., expression, geometry, and pose, via monocular 3D face reconstruction. Next, a recurrent network is introduced to translate source audio into expression parameters that are primarily related to the audio content. The audio-translated expression parameters are then used to synthesize a photo-realistic human subject in each video frame, with the movement of the mouth regions precisely mapped to the source audio. The geometry and pose parameters of the target human portrait are retained, therefore preserving the con-text of the original video footage. Finally, we introduce a novel video rendering network and a dynamic programming method to construct a temporally coherent and photo-realistic video. Extensive experiments demonstrate the superiority of our method over existing approaches. Our method is end-to-end learnable and robust to voice variations in the source audio. | Withdrawn |
ICLR.cc/2022/Conference | Learning Universal User Representations via Self-Supervised Lifelong Behaviors Modeling | Universal user representation is an important research topic in industry, and is widely used in diverse downstream user analysis tasks, such as user profiling and user preference prediction. With the rapid development of Internet service platforms, extremely long user behavior sequences have been accumulated. However, existing researches have little ability to model universal user representation based on lifelong behavior sequences since user registration. In this study, we propose a novel framework called Lifelong User Representation Model (LURM) to tackle this challenge. Specifically, LURM consists of two cascaded sub-models: (i) Bag of Interests (BoI) encodes user behaviors in any time period into a sparse vector with super-high dimension (eg. 10^5); (ii) Self-supervised Multi-anchor Encoder Network (SMEN) maps sequences of BoI features to multiple low-dimensional user representations by contrastive learning. SMEN achieves almost lossless dimensionality reduction with the main help of a novel multi-anchor module which can learn different aspects of user preferences. Experiments on several benchmark datasets show that our approach can outperform state-of-the-art unsupervised representation methods in downstream tasks. | Reject |
ICLR.cc/2021/Conference | Improving Local Effectiveness for Global Robustness Training | Despite its increasing popularity, deep neural networks are easily fooled. To alleviate this deficiency, researchers are actively developing new training strategies, which encourage models that are robust to small input perturbations. Several successful robust training methods have been proposed. However, many of them rely on strong adversaries, which can be prohibitively expensive to generate when the input dimension is high and the model structure is complicated. We adopt a new perspective on robustness and propose a novel training algorithm that allows a more effective use of adversaries. Our method improves the model robustness at each local ball centered around an adversary and then, by combining these local balls through a global term, achieves overall robustness. We demonstrate that, by maximizing the use of adversaries via focusing on local balls, we achieve high robust accuracy with weak adversaries. Specifically, our method reaches a similar robust accuracy level to the state of the art approaches trained on strong adversaries on MNIST, CIFAR-10 and CIFAR-100. As a result, the overall training time is reduced. Furthermore, when trained with strong adversaries, our method matches with the current state of the art on MNIST and outperforms them on CIFAR-10 and CIFAR-100. | Reject |
ICLR.cc/2021/Conference | Data augmentation for deep learning based accelerated MRI reconstruction | Deep neural networks have emerged as very successful tools for image restoration and reconstruction tasks. These networks are often trained end-to-end to directly reconstruct an image from a noisy or corrupted measurement of that image. To achieve state-of-the-art performance, training on large and diverse sets of images is considered critical. However, it is often difficult and/or expensive to collect large amounts of training images. Inspired by the success of Data Augmentation (DA) for classification problems, in this paper, we propose a pipeline for data augmentation for image reconstruction tasks arising in medical imaging and explore its effectiveness at reducing the required training data in a variety of settings. We focus on accelerated magnetic resonance imaging, where the goal is to reconstruct an image from a few under-sampled linear measurements. Our DA pipeline is specifically designed to utilize the invariances present in medical imaging measurements as naive DA strategies that neglect the physics of the problem fail. We demonstrate the effectiveness of our data augmentation pipeline by showing that for some problem regimes, DA can achieve comparable performance to the state of the art on the FastMRI dataset while using significantly fewer training data. Specifically, for 8-fold acceleration we achieve performance comparable to the state of the art with only $10\%$ of the training data for multi-coil reconstruction and with only $33\%$ of the training data for single-coil reconstruction. Our findings show that in the low-data regime DA is beneficial, whereas in the high-data regime it has diminishing returns. | Reject |
ICLR.cc/2021/Conference | GAN2GAN: Generative Noise Learning for Blind Denoising with Single Noisy Images | We tackle a challenging blind image denoising problem, in which only single distinct noisy images are available for training a denoiser, and no information about noise is known, except for it being zero-mean, additive, and independent of the clean image. In such a setting, which often occurs in practice, it is not possible to train a denoiser with the standard discriminative training or with the recently developed Noise2Noise (N2N) training; the former requires the underlying clean image for the given noisy image, and the latter requires two independently realized noisy image pair for a clean image. To that end, we propose GAN2GAN (Generated-Artificial-Noise to Generated-Artificial-Noise) method that first learns a generative model that can 1) simulate the noise in the given noisy images and 2) generate a rough, noisy estimates of the clean images, then 3) iteratively trains a denoiser with subsequently synthesized noisy image pairs (as in N2N), obtained from the generative model. In results, we show the denoiser trained with our GAN2GAN achieves an impressive denoising performance on both synthetic and real-world datasets for the blind denoising setting; it almost approaches the performance of the standard discriminatively-trained or N2N-trained models that have more information than ours, and it significantly outperforms the recent baseline for the same setting, \textit{e.g.}, Noise2Void, and a more conventional yet strong one, BM3D. The official code of our method is available at https://github.com/csm9493/GAN2GAN. | Accept (Poster) |
ICLR.cc/2020/Conference | Weighted Empirical Risk Minimization: Transfer Learning based on Importance Sampling | We consider statistical learning problems, when the distribution $P'$ of the training observations $Z'_1,\; \ldots,\; Z'_n$ differs from the distribution $P$ involved in the risk one seeks to minimize (referred to as the \textit{test distribution}) but is still defined on the same measurable space as $P$ and dominates it. In the unrealistic case where the likelihood ratio $\Phi(z)=dP/dP'(z)$ is known, one may straightforwardly extends the Empirical Risk Minimization (ERM) approach to this specific \textit{transfer learning} setup using the same idea as that behind Importance Sampling, by minimizing a weighted version of the empirical risk functional computed from the 'biased' training data $Z'_i$ with weights $\Phi(Z'_i)$. Although the \textit{importance function} $\Phi(z)$ is generally unknown in practice, we show that, in various situations frequently encountered in practice, it takes a simple form and can be directly estimated from the $Z'_i$'s and some auxiliary information on the statistical population $P$. By means of linearization techniques, we then prove that the generalization capacity of the approach aforementioned is preserved when plugging the resulting estimates of the $\Phi(Z'_i)$'s into the weighted empirical risk. Beyond these theoretical guarantees, numerical results provide strong empirical evidence of the relevance of the approach promoted in this article. | Reject |
ICLR.cc/2018/Conference | Quadrature-based features for kernel approximation | We consider the problem of improving kernel approximation via feature maps. These maps arise as Monte Carlo approximation to integral representations of kernel functions and scale up kernel methods for larger datasets. We propose to use more efficient numerical integration technique to obtain better estimates of the integrals compared to the state-of-the-art methods. Our approach allows to use information about the integrand to enhance approximation and facilitates fast computations. We derive the convergence behavior and conduct an extensive empirical study that supports our hypothesis. | Reject |
ICLR.cc/2023/Conference | AIA: learn to design greedy algorithm for NP-complete problems using neural networks | Algorithm design is an art that heavily requires intuition and expertise of the human designers as well as insights into the problems under consideration. In particular, the design of greedy-selection rules, the core of greedy algorithms, is usually a great challenge to designer: it is relatively easy to understand a greedy algorithm while it is always difficult to find out an effective greedy-selection rule. In the study, we present an approach, called AIA, to learn algorithm design with the aid of neural networks. We consider the minimum weighted set cover (SC) problem, one of the NP-hard problems, as an representative example. Initially, we formulate a given weighted SC problem as an 0-1 integer linear program (ILP): each variable $x_i$ has two options, i.e., $x_i=0$, which denotes abandon of the set $s_i$, and $x_i = 1$, which denotes selection of $s_i$. Each option of a variable leads to a sub-problem with respect to the original ILP problem. Next, we design a generic search framework to find the optimal solution to the ILP problem. At each search step, the value of a variable is determined with the aid of neural networks. The key of our neural network is the loss function: the original ILP problem and the sub-problems generated by assigning a variable $x_i$ should satisfy the Bellman-Ford equation, and the dissatisfication of the Bellman-Ford equation is evaluated and used as loss function of our neural network. The trained neural network is used as greedy-selection rule. Experimental results on representative instances suggest that using the NN-based greedy selection rule, we can successfully find the optimal solutions. More importantly, the NN-based greedy-selection rule outperform the outstanding Chavatal greedy algorithm, which was designed by human expert. The basic idea of our approach can be readily extended without significant modification to design greedy algorithm for other NP-hard problems. | Reject |
ICLR.cc/2022/Conference | Implicit Bias of MSE Gradient Optimization in Underparameterized Neural Networks | We study the dynamics of a neural network in function space when optimizing the mean squared error via gradient flow. We show that in the underparameterized regime the network learns eigenfunctions of an integral operator $T_K$ determined by the Neural Tangent Kernel at rates corresponding to their eigenvalues. For example, for uniformly distributed data on the sphere $S^{d - 1}$ and rotation invariant weight distributions, the eigenfunctions of $T_K$ are the spherical harmonics. Our results can be understood as describing a spectral bias in the underparameterized regime. The proofs use the concept of ``Damped Deviations'' where deviations of the NTK matter less for eigendirections with large eigenvalues. Aside from the underparameterized regime, the damped deviations point-of-view allows us to extend certain results in the literature in the overparameterized setting. | Accept (Poster) |
ICLR.cc/2023/Conference | Label-free Concept Bottleneck Models | Concept bottleneck models (CBM) are a popular way of creating more interpretable neural networks by having hidden layer neurons correspond to human-understandable concepts. However, existing CBMs and their variants have two crucial limitations: first, they need to collect labeled data for each of the predefined concepts, which is time consuming and labor intensive; second, the accuracy of a CBM is often significantly lower than that of a standard neural network, especially on more complex datasets. This poor performance creates a barrier for adopting CBMs in practical real world applications. Motivated by these challenges, we propose Label-free CBM which is a novel framework to transform any neural network into an interpretable CBM without labeled concept data, while retaining a high accuracy. Our Label-free CBM has many advantages, it is: scalable - we present the first CBM scaled to ImageNet, efficient - creating a CBM takes only a few hours even for very large datasets, and automated - training it for a new dataset requires minimal human effort. Our code is available at https://github.com/Trustworthy-ML-Lab/Label-free-CBM. | Accept: poster |
ICLR.cc/2021/Conference | Not All Memories are Created Equal: Learning to Expire | Attention mechanisms have shown promising results in sequence modeling tasks that require long-term memory. Recent work has investigated mechanisms to reduce the computational cost of preserving and storing the memories. However, not all content in the past is equally important to remember. We propose Expire-Span, a method that learns to retain the most important information and
expire the irrelevant information. This enables Transformers to scale to attend to tens of thousands of previous timesteps efficiently, as not all hidden states from previous timesteps are preserved. We demonstrate that Expire-Span can help models identify and retain critical information and show it can achieve state of the art results on long-context language modeling, reinforcement learning, and algorithmic tasks. Finally, we show that Expire-Span can scale to memories that are tens of thousands in size, which is helpful on incredibly long context tasks such as character-level PG-19 and a frame-by-frame moving objects task.
| Reject |
ICLR.cc/2023/Conference | Multi Task Learning of Different Class Label Representations for Stronger Models | We find that the way in which class labels are represented can have a powerful effect on how well models trained on them learn. In classification, the standard way of representing class labels is as one-hot vectors. We present a new way of representing class labels called Binary Labels, where each class label is a large binary vector. We further introduce a new paradigm, multi task learning on different label representations. We train a network on two tasks. The main task is to classify images based on their one-hot label, and the auxiliary task is to classify images based on their Binary Label. We show that networks trained on both tasks have many advantages, including higher accuracy across a wide variety of datasets and architectures, both when trained from scratch and when using transfer learning. Networks trained on both tasks are also much more effective when training data is limited, and seem to do especially well on more challenging problems. | Reject |
ICLR.cc/2022/Conference | Shap-CAM: Visual Explanations for Convolutional Neural Networks based on Shapley Value | Explaining deep convolutional neural networks has been recently drawing increasing attention since it helps to understand the networks' internal operations and why they make certain decisions. Saliency maps, which emphasize salient regions largely connected to the network's decision-making, are one of the most common ways for visualizing and analyzing deep networks in the computer vision community. However, saliency maps generated by existing methods cannot represent authentic information in images due to the unproven proposals about the weights of activation maps which lack solid theoretical foundation and fail to consider the relations between each pixels. In this paper, we develop a novel post-hoc visual explanation method called Shap-CAM based on class activation mapping. Unlike previous class activation mapping based approaches, Shap-CAM gets rid of the dependence on gradients by obtaining the importance of each pixels through Shapley value. We demonstrate that Shap-CAM achieves better visual performance and fairness for interpreting the decision making process. Our approach outperforms previous methods on both recognition and localization tasks. | Withdrawn |
ICLR.cc/2018/Conference | CausalGAN: Learning Causal Implicit Generative Models with Adversarial Training | We introduce causal implicit generative models (CiGMs): models that allow sampling from not only the true observational but also the true interventional distributions. We show that adversarial training can be used to learn a CiGM, if the generator architecture is structured based on a given causal graph. We consider the application of conditional and interventional sampling of face images with binary feature labels, such as mustache, young. We preserve the dependency structure between the labels with a given causal graph. We devise a two-stage procedure for learning a CiGM over the labels and the image. First we train a CiGM over the binary labels using a Wasserstein GAN where the generator neural network is consistent with the causal graph between the labels. Later, we combine this with a conditional GAN to generate images conditioned on the binary labels. We propose two new conditional GAN architectures: CausalGAN and CausalBEGAN. We show that the optimal generator of the CausalGAN, given the labels, samples from the image distributions conditioned on these labels. The conditional GAN combined with a trained CiGM for the labels is then a CiGM over the labels and the generated image. We show that the proposed architectures can be used to sample from observational and interventional image distributions, even for interventions which do not naturally occur in the dataset. | Accept (Poster) |
ICLR.cc/2021/Conference | Transferring Inductive Biases through Knowledge Distillation | Having the right inductive biases can be crucial in many tasks or scenarios where data or computing resources are a limiting factor, or where training data is not perfectly representative of the conditions at test time. However, defining, designing, and efficiently adapting inductive biases is not necessarily straightforward. Inductive biases of a model affect its generalisation behaviour and influence the solution it converges to from different aspects. In this paper, we investigate the power of knowledge distillation in transferring the effects of inductive biases of a teacher model to a student model, when they have different architectures.
We consider different families of models: LSTMs vs. Transformers and CNNs vs. MLPs, in the context of tasks and scenarios with linguistics and vision applications, where having the right inductive biases is critical. We train our models in different setups: no knowledge distillation, self-distillation, and distillation using a teacher with a better inductive bias for the task at hand. We show that in the later setup, compared to no distillation and self-distillation, we can not only improve the performance of the students, but also the solutions they converge become similar to their teachers with respect to a wide range of properties, including different task-specific performance metrics, per sample behaviour of the models, representational similarity and how the representational space of the models evolve during training, performance on out-of-distribution datasets, confidence calibration, and finally whether the converged solutions fall within the same basins of attractions. | Reject |
ICLR.cc/2020/Conference | Task-agnostic Continual Learning via Growing Long-Term Memory Networks | As our experience shows, humans can learn and deploy a myriad of different skills to tackle the situations they encounter daily. Neural networks, in contrast, have a fixed memory capacity that prevents them from learning more than a few sets of skills before starting to forget them.
In this work, we make a step to bridge neural networks with human-like learning capabilities. For this, we propose a model with a growing and open-bounded memory capacity that can be accessed based on the model’s current demands. To test this system, we introduce a continual learning task based on language modelling where the model is exposed to multiple languages and domains in sequence, without providing any explicit signal on the type of input it is currently dealing with. The proposed system exhibits improved adaptation skills in that it can recover faster than comparable baselines after a switch in the input language or domain. | Withdrawn |
ICLR.cc/2020/Conference | SCALOR: Generative World Models with Scalable Object Representations | Scalability in terms of object density in a scene is a primary challenge in unsupervised sequential object-oriented representation learning. Most of the previous models have been shown to work only on scenes with a few objects. In this paper, we propose SCALOR, a probabilistic generative world model for learning SCALable Object-oriented Representation of a video. With the proposed spatially parallel attention and proposal-rejection mechanisms, SCALOR can deal with orders of magnitude larger numbers of objects compared to the previous state-of-the-art models. Additionally, we introduce a background module that allows SCALOR to model complex dynamic backgrounds as well as many foreground objects in the scene. We demonstrate that SCALOR can deal with crowded scenes containing up to a hundred objects while jointly modeling complex dynamic backgrounds. Importantly, SCALOR is the first unsupervised object representation model shown to work for natural scenes containing several tens of moving objects. | Accept (Poster) |
ICLR.cc/2020/Conference | Improved Training Techniques for Online Neural Machine Translation | Neural sequence-to-sequence models are at the basis of state-of-the-art solutions for sequential prediction problems such as machine translation and speech recognition. The models typically assume that the entire input is available when starting target generation. In some applications, however, it is desirable to start the decoding process before the entire input is available, e.g. to reduce the latency in automatic speech recognition. We consider state-of-the-art wait-k decoders, that first read k tokens from the source and then alternate between reading tokens from the input and writing to the output. We investigate the sensitivity of such models to the value of k that is used during training and when deploying the model, and the effect of updating the hidden states in transformer models as new source tokens are read. We experiment with German-English translation on the IWSLT14 dataset and the larger WMT15 dataset. Our results significantly improve over earlier state-of-the-art results for German-English translation on the WMT15 dataset across different latency levels. | Reject |
ICLR.cc/2022/Conference | Coordination Among Neural Modules Through a Shared Global Workspace | Deep learning has seen a movement away from representing examples with a monolithic hidden state towards a richly structured state. For example, Transformers segment by position, and object-centric architectures decompose images into entities. In all these architectures, interactions between different elements are modeled via pairwise interactions: Transformers make use of self-attention to incorporate information from other positions and object-centric architectures make use of graph neural networks to model interactions among entities. We consider how to improve on pairwise interactions in terms of global coordination and a coherent, integrated representation that can be used for downstream tasks. In cognitive science, a global workspace architecture has been proposed in which functionally specialized components share information through a common, bandwidth-limited communication channel. We explore the use of such a communication channel in the context of deep learning for modeling the structure of complex environments. The proposed method includes a shared workspace through which communication among different specialist modules takes place but due to limits on the communication bandwidth, specialist modules must compete for access. We show that capacity limitations have a rational basis in that (1) they encourage specialization and compositionality and (2) they facilitate the synchronization of otherwise independent specialists.
| Accept (Oral) |
ICLR.cc/2022/Conference | MemREIN: Rein the Domain Shift for Cross-Domain Few-Shot Learning | Few-shot learning aims to enable models generalize to new categories (query instances) with only limited labeled samples (support instances) from each category. Metric-based mechanism is a promising direction which compares feature embeddings via different metrics. However, it always fail to generalize to unseen domains due to the considerable domain gap challenge. In this paper, we propose a novel framework, MemREIN, which considers Memorized, Restitution, and Instance Normalization for cross-domain few-shot learning. Specifically, an instance normalization algorithm is explored to alleviate feature dissimilarity, which provides the initial model generalization ability. However, naively normalizing the feature would lose fine-grained discriminative knowledge between different classes. To this end, a memorized module is further proposed to separate the most refined knowledge and remember it. Then, a restitution module is utilized to restitute the discrimination ability from the learned knowledge. A novel reverse contrastive learning strategy is proposed to stabilize the distillation process. Extensive experiments on five popular benchmark datasets demonstrate that MemREIN well addresses the domain shift challenge, and significantly improves the performance up to $16.37\%$ compared with state-of-the-art baselines. | Withdrawn |
ICLR.cc/2023/Conference | Normalizing Flows for Interventional Density Estimation | Existing machine learning methods for causal inference usually estimate quantities expressed via the mean of potential outcomes (e.g., average treatment effect). However, such quantities do not capture the full information about the distribution of potential outcomes. In this work, we estimate the density of potential outcomes after interventions from observational data. For this, we propose a novel, fully-parametric deep learning method called Interventional Normalizing Flows. Specifically, we combine two normalizing flows, namely (i) a teacher flow for estimating nuisance parameters and (ii) a student flow for a parametric estimation of the density of potential outcomes. We further develop a tractable optimization objective via a one-step bias correction for an efficient and doubly robust estimation of the student flow parameters. As a result our Interventional Normalizing Flows offer a properly normalized density estimator. Across various experiments, we demonstrate that our Interventional Normalizing Flows are expressive and highly effective, and scale well with both sample size and high-dimensional confounding. To the best of our knowledge, our Interventional Normalizing Flows are the first fully-parametric, deep learning method for density estimation of potential outcomes. | Reject |
ICLR.cc/2022/Conference | Ancestral protein sequence reconstruction using a tree-structured Ornstein-Uhlenbeck variational autoencoder | We introduce a deep generative model for representation learning of biological sequences that, unlike existing models, explicitly represents the evolutionary process. The model makes use of a tree-structured Ornstein-Uhlenbeck process, obtained from a given phylogenetic tree, as an informative prior for a variational autoencoder. We show the model performs well on the task of ancestral sequence reconstruction of single protein families. Our results and ablation studies indicate that the explicit representation of evolution using a suitable tree-structured prior has the potential to improve representation learning of biological sequences considerably. Finally, we briefly discuss extensions of the model to genomic-scale data sets and the case of a latent phylogenetic tree. | Accept (Poster) |
ICLR.cc/2019/Conference | An Efficient and Margin-Approaching Zero-Confidence Adversarial Attack | There are two major paradigms of white-box adversarial attacks that attempt to impose input perturbations. The first paradigm, called the fix-perturbation attack, crafts adversarial samples within a given perturbation level. The second paradigm, called the zero-confidence attack, finds the smallest perturbation needed to cause misclassification, also known as the margin of an input feature. While the former paradigm is well-resolved, the latter is not. Existing zero-confidence attacks either introduce significant approximation errors, or are too time-consuming. We therefore propose MarginAttack, a zero-confidence attack framework that is able to compute the margin with improved accuracy and efficiency. Our experiments show that MarginAttack is able to compute a smaller margin than the state-of-the-art zero-confidence attacks, and matches the state-of-the-art fix-perturbation attacks. In addition, it runs significantly faster than the Carlini-Wagner attack, currently the most accurate zero-confidence attack algorithm. | Reject |
ICLR.cc/2020/Conference | Multi-agent Reinforcement Learning for Networked System Control | This paper considers multi-agent reinforcement learning (MARL) in networked system control. Specifically, each agent learns a decentralized control policy based on local observations and messages from connected neighbors. We formulate such a networked MARL (NMARL) problem as a spatiotemporal Markov decision process and introduce a spatial discount factor to stabilize the training of each local agent. Further, we propose a new differentiable communication protocol, called NeurComm, to reduce information loss and non-stationarity in NMARL. Based on experiments in realistic NMARL scenarios of adaptive traffic signal control and cooperative adaptive cruise control, an appropriate spatial discount factor effectively enhances the learning curves of non-communicative MARL algorithms, while NeurComm outperforms existing communication protocols in both learning efficiency and control performance. | Accept (Poster) |
ICLR.cc/2023/Conference | MixPro: Data Augmentation with MaskMix and Progressive Attention Labeling for Vision Transformer | The recently proposed data augmentation TransMix employs attention labels to help visual transformers (ViT) achieve better robustness and performance. However, TransMix is deficient in two aspects: 1) The image cropping method of TransMix may not be suitable for vision transformer. 2) At the early stage of training, the model produces unreliable attention maps. TransMix uses unreliable attention maps to compute mixed attention labels that can affect the model. To address the aforementioned issues, we propose MaskMix and Progressive Attention Labeling (PAL) in image and label space, respectively. In detail, from the perspective of image space, we design MaskMix, which mixes two images based on a patch-like grid mask. In particular, the size of each mask patch is adjustable and is a multiple of the image patch size, which ensures each image patch comes from only one image and contains more global contents. From the perspective of label space, we design PAL, which utilizes a progressive factor to dynamically re-weight the attention weights of the mixed attention label. Finally, we combine MaskMix and Progressive Attention Labeling as our new data augmentation method, named MixPro. The experimental results show that our method can improve various ViT-based models at scales on ImageNet classification (73.8% top-1 accuracy based on DeiT-T for 300 epochs). After being pre-trained with MixPro on ImageNet, the ViT-based models also demonstrate better transferability to semantic segmentation, object detection, and instance segmentation. Furthermore, compared to TransMix, MixPro also shows stronger robustness on several benchmarks. | Accept: poster |
ICLR.cc/2019/Conference | SGD Converges to Global Minimum in Deep Learning via Star-convex Path | Stochastic gradient descent (SGD) has been found to be surprisingly effective in training a variety of deep neural networks. However, there is still a lack of understanding on how and why SGD can train these complex networks towards a global minimum. In this study, we establish the convergence of SGD to a global minimum for nonconvex optimization problems that are commonly encountered in neural network training. Our argument exploits the following two important properties: 1) the training loss can achieve zero value (approximately), which has been widely observed in deep learning; 2) SGD follows a star-convex path, which is verified by various experiments in this paper. In such a context, our analysis shows that SGD, although has long been considered as a randomized algorithm, converges in an intrinsically deterministic manner to a global minimum. | Accept (Poster) |
ICLR.cc/2023/Conference | Hierarchies of Reward Machines | Reward machines (RMs) are a recent formalism for representing the reward function of a reinforcement learning task through a finite-state machine whose edges encode landmarks of the task using high-level events. The structure of RMs enables the decomposition of a task into simpler and independently solvable subtasks that help tackle long-horizon and/or sparse reward tasks. We propose a formalism for further abstracting the subtask structure by endowing an RM with the ability to call other RMs, thus composing a hierarchy of RMs (HRM). We exploit HRMs by treating each call to an RM as an independently solvable subtask using the options framework, and describe a curriculum-based method to learn HRMs from traces observed by the agent. Our experiments reveal that exploiting a handcrafted HRM leads to faster convergence than with a flat HRM, and that learning an HRM remains feasible in cases where its equivalent flat representation is not. | Reject |
ICLR.cc/2022/Conference | An Improved Composite Functional Gradient Learning by Wasserstein Regularization for Generative adversarial networks | Generative adversarial networks (GANs) are usually trained by a minimax game which is notoriously and empirically known to be unstable. Recently, a totally new methodology called
Composite Functional Gradient Learning (CFG) provides an alternative
theoretical foundation for training GANs more stablely by employing
a strong discriminator with logistic regression and functional gradient learning for the generator.
However, the discriminator using logistic regression from the CFG framework is gradually
hard to discriminate between real and fake images while the training steps go on.
To address this problem, our key idea and contribution are to introduce
the Wasserstein distance regularization into the CFG framework for the discriminator. This gives us a novel
improved CFG formulation with more competitive generate image quality. In particular, we provide an intuitive explanation using logistic regression with Wasserstein regularization. The method helps to enhance the model gradients
in training GANs to archives better image quality. Empirically, we compare our improved CFG with the original
version. We show that the standard CFG is easy to stick into mode collapse problem, while our improved CFG works much better
thanks to the newly added Wasserstein distance regularization. We conduct extensive
experiments for image generation on different benchmarks, and it shows
the efficacy of our improved CFG method.
| Withdrawn |
ICLR.cc/2023/Conference | LSTM-BASED-AUTO-BI-LSTM for Remaining Useful Life (RUL) Prediction: the first round of test results | The Remaining Useful Life (RUL) is one of the most critical indicators to detect a component’s failure before it effectively occurs. It can be predicted by historical data or direct data extraction by adopting model-based, data-driven, or hybrid methodologies. Data-driven methods have mainly used Machine Learning (ML) approaches, despite several studies still pointing out different challenges in this sense. For instance, traditional ML methods cannot extract features directly from time series depending, in some cases, on the prior knowledge of the system. In this context, this work proposes a DL-based approach called LSTM-based-AUTO-Bi-LSTM. It ensembles an LSTM-based autoencoder to automatically perform feature engineering (instead of manually) with Bidirectional Long Short-Term Memory (Bi-LSTM) to predict RUL. We have tested the model using the Turbofan Engine Degradation Simulation Dataset (FD001), an open dataset. It was generated from the Commercial Modular Aero-Propulsion System Simulation (C-MAPSS) from the Prognostics Center of Excellence (PcoE), from the National Aeronautics and Space Administration (NASA). The objective is to release the first round of analytical results and statistical visualisations of the model application, which will guide us in future improvements. | Reject |
ICLR.cc/2022/Conference | Information-theoretic stochastic contrastive conditional GAN: InfoSCC-GAN | Conditional generation is a subclass of generative problems when the output of generation is conditioned by a class attributes’ information. In this paper, we present a new stochastic contrastive conditional generative adversarial network (InfoSCC-GAN) with explorable latent space. The InfoSCC-GAN architecture is based on an unsupervised contrastive encoder built on the InfoNCE paradigm, attributes' classifier, and stochastic EigenGAN generator.
We propose two approaches for selecting the class attributes: external attributes from the dataset annotations and internal attributes from the clustered latent space of the encoder. We propose a novel training method based on a generator regularization using external or internal attributes every $n$-th iteration using the pre-trained contrastive encoder and pre-trained attributes’ classifier. The proposed InfoSCC-GAN is derived from an information-theoretic formulation of mutual information maximization between the input data and latent space representation for the encoder and the latent space and generated data for the decoder. Thus, we demonstrate a link between the training objective functions and the above information-theoretic formulation. The experimental results show that InfoSCC-GAN outperforms vanilla EigenGAN in image generation on several popular datasets, yet providing an interpretable latent space. In addition, we investigate the impact of regularization techniques and each part of the system by performing an ablation study. Finally, we demonstrate that thanks to the stochastic EigenGAN generator, the proposed framework enjoys a truly stochastic generation in contrast to vanilla deterministic GANs yet with the independent training of an encoder, a classifier, and a generator.
The code, supplementary materials, and demos are available \url{https://anonymous.4open.science/r/InfoSCC-GAN-D113} | Reject |
ICLR.cc/2020/Conference | Cross-Dimensional Self-Attention for Multivariate, Geo-tagged Time Series Imputation | Many real-world applications involve multivariate, geo-tagged time series data: at each location, multiple sensors record corresponding measurements. For example, air quality monitoring system records PM2.5, CO, etc. The resulting time-series data often has missing values due to device outages or communication errors. In order to impute the missing values, state-of-the-art methods are built on Recurrent Neural Networks (RNN), which process each time stamp sequentially, prohibiting the direct modeling of the relationship between distant time stamps. Recently, the self-attention mechanism has been proposed for sequence modeling tasks such as machine translation, significantly outperforming RNN because the relationship between each two time stamps can be modeled explicitly. In this paper, we are the first to adapt the self-attention mechanism for multivariate, geo-tagged time series data. In order to jointly capture the self-attention across different dimensions (i.e. time, location and sensor measurements) while keep the size of attention maps reasonable, we propose a novel approach called Cross-Dimensional Self-Attention (CDSA) to process each dimension sequentially, yet in an order-independent manner. On three real-world datasets, including one our newly collected NYC-traffic dataset, extensive experiments demonstrate the superiority of our approach compared to state-of-the-art methods for both imputation and forecasting tasks.
| Reject |
ICLR.cc/2020/Conference | Neural Network Branching for Neural Network Verification | Formal verification of neural networks is essential for their deployment in safety-critical areas. Many available formal verification methods have been shown to be instances of a unified Branch and Bound (BaB) formulation. We propose a novel framework for designing an effective branching strategy for BaB. Specifically, we learn a graph neural network (GNN) to imitate the strong branching heuristic behaviour. Our framework differs from previous methods for learning to branch in two main aspects. Firstly, our framework directly treats the neural network we want to verify as a graph input for the GNN. Secondly, we develop an intuitive forward and backward embedding update schedule. Empirically, our framework achieves roughly $50\%$ reduction in both the number of branches and the time required for verification on various convolutional networks when compared to the best available hand-designed branching strategy. In addition, we show that our GNN model enjoys both horizontal and vertical transferability. Horizontally, the model trained on easy properties performs well on properties of increased difficulty levels. Vertically, the model trained on small neural networks achieves similar performance on large neural networks. | Accept (Talk) |
ICLR.cc/2023/Conference | Tailoring Language Generation Models under Total Variation Distance | The standard paradigm of neural language generation adopts maximum likelihood estimation (MLE) as the optimizing method. From a distributional view, MLE in fact minimizes the Kullback-Leibler divergence (KLD) between the distribution of the real data and that of the model. However, this approach forces the model to distribute non-zero (sometimes large) probability mass to all training samples regardless of their quality. Moreover, in the attempt to cover the low-probability regions in the data distribution, the model systematically overestimates the probability of corrupted text sequences, which we conjecture is one of the main reasons for text degeneration during autoregressive decoding. To remedy this problem, we leverage the total variation distance (TVD) with its robustness to outliers, and develop practical bounds to apply it to language generation. Then, we introduce the TaiLr objective that balances the tradeoff of estimating TVD. Intuitively, TaiLr downweights real data samples that have low model probabilities with tunable penalization intensity. Experimental results show that our method alleviates the overestimation of degenerated sequences without sacrificing diversity and improves generation quality on a wide range of text generation tasks. | Accept: notable-top-5% |
ICLR.cc/2020/Conference | On the implicit minimization of alternative loss functions when training deep networks | Understanding the implicit bias of optimization algorithms is important in order to improve generalization of neural networks. One approach to try to exploit such understanding would be to then make the bias explicit in the loss function. Conversely, an interesting approach to gain more insights into the implicit bias could be to study how different loss functions are being implicitly minimized when training the network. In this work, we concentrate our study on the inductive bias occurring when minimizing the cross-entropy loss with different batch sizes and learning rates. We investigate how three loss functions are being implicitly minimized during training. These three loss functions are the Hinge loss with different margins, the cross-entropy loss with different temperatures and a newly introduced Gcdf loss with different standard deviations. This Gcdf loss establishes a connection between a sharpness measure for the 0−1 loss and margin based loss functions. We find that a common behavior is emerging for all the loss functions considered. | Reject |
ICLR.cc/2022/Conference | Connecting Graph Convolution and Graph PCA | Graph convolution operator of the GCN model is originally motivated from a localized first-order approximation of spectral graph convolutions. This work stands on a different view; establishing a mathematical connection between graph convolution and graph-regularized PCA (GPCA). Based on this connection, the GCN architecture, shaped by stacking graph convolution layers, shares a close relationship with stacking GPCA. We empirically demonstrate that the unsupervised embeddings by GPCA paired with a 1- or 2-layer MLP achieves similar or even better performance than many sophisticated baselines on semi-supervised node classification tasks across five datasets including Open Graph Benchmark. This suggests that the prowess of graph convolution is driven by graph based regularization. In addition, we extend GPCA to the (semi-)supervised setting and show that it is equivalent to GPCA on a graph extended with “ghost” edges between nodes of the same label. Finally, we capitalize on the discovered relationship to design an effective initialization strategy based on stacking GPCA, enabling GCN to converge faster and achieve robust performance at large number of layers. | Reject |
ICLR.cc/2022/Conference | Towards Unknown-aware Deep Q-Learning | Deep reinforcement learning (RL) has achieved remarkable success in known environments where the agents are trained, yet the agents do not necessarily know what they don’t know. In particular, RL agents deployed in the open world are naturally subject to environmental shifts and encounter unknown out-of-distribution (OOD) states---i.e., states from outside the training environment. Currently, the study of handling OOD states in the RL environment remains underexplored. This paper bridges this critical gap by proposing and exploring an unknown-aware RL framework, which improves the safety and reliability of deep Q-learning. Our key idea is to regularize the training of Q-learning so that OOD states will have higher OOD uncertainty, while in-distribution states will have lower OOD uncertainty; therefore making them distinguishable. This is in contrast with vanilla Q-learning which does not take into account unknowns during training. Furthermore, we provide theoretical guarantees that our method can improve OOD uncertainty estimation while ensuring the convergence performance of the in-distribution environment. Empirically, we demonstrate state-of-the-art performance on six diverse environments, achieving near-optimal OOD detection performance. | Withdrawn |
ICLR.cc/2022/Conference | A Variance Principle Explains why Dropout Finds Flatter Minima | Although dropout has achieved great success in deep learning, little is known about how it helps the training find a good generalization solution in the high-dimensional parameter space. In this work, we show that the training with dropout finds the neural network with a flatter minimum compared with standard gradient descent training. We further study the underlying mechanism of why dropout finds flatter minima through experiments. We propose a Variance Principle that the variance of a noise is larger at the sharper direction of the loss landscape. Existing works show that SGD satisfies the variance principle, which leads the training to flatter minima. Our work show that the noise induced by the dropout also satisfies the variance principle that explains why dropout finds flatter minima. In general, our work points out that the variance principle is an important similarity between dropout and SGD that lead the training to find flatter minima and obtain good generalization. | Reject |
ICLR.cc/2021/Conference | Set Prediction without Imposing Structure as Conditional Density Estimation | Set prediction is about learning to predict a collection of unordered variables with unknown interrelations. Training such models with set losses imposes the structure of a metric space over sets. We focus on stochastic and underdefined cases, where an incorrectly chosen loss function leads to implausible predictions. Example tasks include conditional point-cloud reconstruction and predicting future states of molecules. In this paper we propose an alternative to training via set losses, by viewing learning as conditional density estimation. Our learning framework fits deep energy-based models and approximates the intractable likelihood with gradient-guided sampling. Furthermore, we propose a stochastically augmented prediction algorithm that enables multiple predictions, reflecting the possible variations in the target set. We empirically demonstrate on a variety of datasets the capability to learn multi-modal densities and produce different plausible predictions. Our approach is competitive with previous set prediction models on standard benchmarks. More importantly, it extends the family of addressable tasks beyond those that have unambiguous predictions. | Accept (Poster) |
ICLR.cc/2018/Conference | Ensemble Robustness and Generalization of Stochastic Deep Learning Algorithms | The question why deep learning algorithms generalize so well has attracted increasing
research interest. However, most of the well-established approaches,
such as hypothesis capacity, stability or sparseness, have not provided complete
explanations (Zhang et al., 2016; Kawaguchi et al., 2017). In this work, we focus
on the robustness approach (Xu & Mannor, 2012), i.e., if the error of a hypothesis
will not change much due to perturbations of its training examples, then it
will also generalize well. As most deep learning algorithms are stochastic (e.g.,
Stochastic Gradient Descent, Dropout, and Bayes-by-backprop), we revisit the robustness
arguments of Xu & Mannor, and introduce a new approach – ensemble
robustness – that concerns the robustness of a population of hypotheses. Through
the lens of ensemble robustness, we reveal that a stochastic learning algorithm can
generalize well as long as its sensitiveness to adversarial perturbations is bounded
in average over training examples. Moreover, an algorithm may be sensitive to
some adversarial examples (Goodfellow et al., 2015) but still generalize well. To
support our claims, we provide extensive simulations for different deep learning
algorithms and different network architectures exhibiting a strong correlation between
ensemble robustness and the ability to generalize. | Invite to Workshop Track |
ICLR.cc/2023/Conference | PatchBlender: A Motion Prior for Video Transformers | Transformers have become one of the dominant architectures in the field of computer vision. However, there are yet several challenges when applying such architectures to video data. Most notably, these models struggle to model the temporal patterns of video data effectively. Directly targeting this issue, we introduce PatchBlender, a learnable blending function that operates over patch embeddings across the temporal dimension of the latent space. We show that our method is successful at enabling vision transformers to encode the temporal component of video data. On Something-Something v2 and MOVi-A, we show that our method improves the performance of a ViT-B. PatchBlender has the advantage of being compatible with almost any Transformer architecture and since it is learnable, the model can adaptively turn on or off the prior. It is also extremely lightweight compute-wise, 0.005% the GFLOPs of a ViT-B. | Reject |
ICLR.cc/2023/Conference | FunkNN: Neural Interpolation for Functional Generation | Can we build continuous generative models which generalize across scales, can be evaluated at any coordinate, admit calculation of exact derivatives, and are conceptually simple? Existing MLP-based architectures generate worse samples than the grid-based generators with favorable convolutional inductive biases. Models that focus on generating images at different scales do better, but employ complex architectures not designed for continuous evaluation of images and derivatives.
We take a signal-processing perspective and treat continuous signal generation as interpolation from samples. Indeed, correctly sampled discrete images contain all information about the low spatial frequencies. The question is then how to extrapolate the spectrum in a data-driven way while meeting the above design criteria. Our answer is FunkNN---a novel convolutional network which learns how to reconstruct continuous images at arbitrary coordinates and can be applied to any image dataset. Combined with a discrete generative model it becomes a functional generator which can act as a prior in continuous ill-posed inverse problems. We show that FunkNN generates high-quality continuous images and exhibits strong out-of-distribution performance thanks to its patch-based design. We further showcase its performance in several stylized inverse problems with exact spatial derivatives. | Accept: poster |
ICLR.cc/2021/Conference | Deformable DETR: Deformable Transformers for End-to-End Object Detection | DETR has been recently proposed to eliminate the need for many hand-designed components in object detection while demonstrating good performance. However, it suffers from slow convergence and limited feature spatial resolution, due to the limitation of Transformer attention modules in processing image feature maps. To mitigate these issues, we proposed Deformable DETR, whose attention modules only attend to a small set of key sampling points around a reference. Deformable DETR can achieve better performance than DETR (especially on small objects) with 10$\times$ less training epochs. Extensive experiments on the COCO benchmark demonstrate the effectiveness of our approach. Code is released at https://github.com/fundamentalvision/Deformable-DETR. | Accept (Oral) |
ICLR.cc/2020/Conference | DropGrad: Gradient Dropout Regularization for Meta-Learning | With the growing attention on learning-to-learn new tasks using only a few examples, meta-learning has been widely used in numerous problems such as few-shot classification, reinforcement learning, and domain generalization. However, meta-learning models are prone to overfitting when there are no sufficient training tasks for the meta-learners to generalize. Although existing approaches such as Dropout are widely used to address the overfitting problem, these methods are typically designed for regularizing models of a single task in supervised training. In this paper, we introduce a simple yet effective method to alleviate the risk of overfitting for gradient-based meta-learning. Specifically, during the gradient-based adaptation stage, we randomly drop the gradient in the inner-loop optimization of each parameter in deep neural networks, such that the augmented gradients improve generalization to new tasks. We present a general form of the proposed gradient dropout regularization and show that this term can be sampled from either the Bernoulli or Gaussian distribution. To validate the proposed method, we conduct extensive experiments and analysis on numerous tasks, demonstrating that the gradient dropout regularization mitigates the overfitting problem and improves the performance upon various gradient-based meta-learning frameworks. | Withdrawn |
ICLR.cc/2020/Conference | Making Efficient Use of Demonstrations to Solve Hard Exploration Problems | This paper introduces R2D3, an agent that makes efficient use of demonstrations to solve hard exploration problems in partially observable environments with highly variable initial conditions. We also introduce a suite of eight tasks that combine these three properties, and show that R2D3 can solve several of the tasks where other state of the art methods (both with and without demonstrations) fail to see even a single successful trajectory after tens of billions of steps of exploration. | Accept (Poster) |
ICLR.cc/2023/Conference | Deja Vu: Continual Model Generalization for Unseen Domains | In real-world applications, deep learning models often run in non-stationary environments where the target data distribution continually shifts over time. There have been numerous domain adaptation (DA) methods in both online and offline modes to improve cross-domain adaptation ability. However, these DA methods typically only provide good performance after a long period of adaptation, and perform poorly on new domains before and during adaptation – in what we call the “Unfamiliar Period”, especially when domain shifts happen suddenly and significantly. On the other hand, domain generalization (DG) methods have been proposed to improve the model generalization ability on unadapted domains. However, existing DG works are ineffective for continually changing domains due to severe catastrophic forgetting of learned knowledge. To overcome these limitations of DA and DG in handling the Unfamiliar Period during continual domain shift, we propose RaTP, a framework that focuses on improving models’ target domain generalization (TDG) capability, while also achieving effective target domain adaptation (TDA) capability right after training on certain domains and forgetting alleviation (FA) capability on past domains. RaTP includes a training-free data augmentation module to prepare data for TDG, a novel pseudo-labeling mechanism to provide reliable supervision for TDA, and a prototype contrastive alignment algorithm to align different domains for achieving TDG, TDA and FA. Extensive experiments on Digits, PACS, and DomainNet demonstrate that RaTP significantly outperforms state-of-the-art works from Continual DA, Source-Free DA, Test-Time/Online DA, Single DG, Multiple DG and Unified DA&DG in TDG, and achieves comparable TDA and FA capabilities. | Accept: poster |
ICLR.cc/2023/Conference | Policy Expansion for Bridging Offline-to-Online Reinforcement Learning | Pre-training with offline data and online fine-tuning using reinforcement learning is a promising strategy for learning control policies by leveraging the best of both worlds in terms of sample efficiency and performance. One natural approach is to initialize the policy for online learning with the one trained offline.
In this work, we introduce a policy expansion scheme for this task. After learning the offline policy, we use it as one candidate policy in a policy set, and further learn another policy that will be responsible for further learning as an expansion to the policy set. The two policies will be composed in an adaptive manner for interacting with the environment. With this approach, the policy previously learned offline is fully retained during online learning, thus mitigating the potential issues such as destroying the useful behaviors of the offline policy in the initial stage of online learning while allowing the offline policy participate in the exploration naturally in an adaptive manner. Moreover, new useful behaviors can potentially be captured by the newly added policy through learning.
Experiments are conducted on a number of tasks and the results demonstrate the effectiveness of the proposed approach. | Accept: poster |
ICLR.cc/2023/Conference | Dynamic-Aware GANs: Time-Series Generation with Handy Self-Supervision | This paper presents Dynamic-Aware GAN (DAGAN) as a data-efficient self-supervised paradigm for time-series data generation.
To support sequential generation with sufficient clues of temporal dynamics, we explicitly model the transition dynamics within the data sequence through differencing, thus refining the vanilla sequence into one with inter-correlated triplets to characterize each time-step.
This localized triplet consistent structure contributes to a self-supervision mechanism, which can provide more aspects of supervision for the overall stepwise dependencies encoded within the training data. Such a handy self-supervision mechanism is simple but can be beneficial especially when a model is presented with limited training data. Based on the insight, we present DAGAN which generalizes the locally regularized triplet consistency to distributional-level via dynamic encoding and joint distribution matching.
Experiments on various synthetic and real-world datasets verify that our model achieves superior generation results with better quality and diversity compared with the state-of-the-art benchmarks, especially when the training data is scarce. Moreover, benefited from the dynamic-conditional and dynamic-consistent design, our DAGAN is capable of generating sequences that present specified dynamics. | Withdrawn |
ICLR.cc/2021/Conference | Dual-Tree Wavelet Packet CNNs for Image Classification | In this paper, we target an important issue of deep convolutional neural networks (CNNs) — the lack of a mathematical understanding of their properties. We present an explicit formalism that is motivated by the similarities between trained CNN kernels and oriented Gabor filters for addressing this problem. The core idea is to constrain the behavior of convolutional layers by splitting them into a succession of wavelet packet decompositions, which are modulated by freely-trained mixture weights. We evaluate our approach with three variants of wavelet decompositions with the AlexNet architecture for image classification as an example. The first variant relies on the separable wavelet packet transform while the other two implement the 2D dual-tree real and complex wavelet packet transforms, taking advantage of their feature extraction properties such as directional selectivity and shift invariance. Our experiments show that we achieve the accuracy rate of standard AlexNet, but with a significantly lower number of parameters, and an interpretation of the network that is grounded in mathematical theory. | Reject |
ICLR.cc/2020/Conference | A Novel Analysis Framework of Lower Complexity Bounds for Finite-Sum Optimization | This paper studies the lower bound complexity for the optimization problem whose objective function is the average of $n$ individual smooth convex functions. We consider the algorithm which gets access to gradient and proximal oracle for each individual component.
For the strongly-convex case, we prove such an algorithm can not reach an $\eps$-suboptimal point in fewer than $\Omega((n+\sqrt{\kappa n})\log(1/\eps))$ iterations, where $\kappa$ is the condition number of the objective function. This lower bound is tighter than previous results and perfectly matches the upper bound of the existing proximal incremental first-order oracle algorithm Point-SAGA.
We develop a novel construction to show the above result, which partitions the tridiagonal matrix of classical examples into $n$ groups to make the problem difficult enough to stochastic algorithms.
This construction is friendly to the analysis of proximal oracle and also could be used in general convex and average smooth cases naturally. | Reject |
ICLR.cc/2018/Conference | Model-based imitation learning from state trajectories | Imitation learning from demonstrations usually relies on learning a policy from trajectories of optimal states and actions. However, in real life expert demonstrations, often the action information is missing and only state trajectories are available. We present a model-based imitation learning method that can learn environment-specific optimal actions only from expert state trajectories. Our proposed method starts with a model-free reinforcement learning algorithm with a heuristic reward signal to sample environment dynamics, which is then used to train the state-transition probability. Subsequently, we learn the optimal actions from expert state trajectories by supervised learning, while back-propagating the error gradients through the modeled environment dynamics. Experimental evaluations show that our proposed method successfully achieves performance similar to (state, action) trajectory-based traditional imitation learning methods even in the absence of action information, with much fewer iterations compared to conventional model-free reinforcement learning methods. We also demonstrate that our method can learn to act from only video demonstrations of expert agent for simple games and can learn to achieve desired performance in less number of iterations. | Reject |
ICLR.cc/2023/Conference | Statistical Theory of Differentially Private Marginal-based Data Synthesis Algorithms | Marginal-based methods achieve promising performance in the synthetic data competition hosted by the National Institute of Standards and Technology (NIST).
To deal with high-dimensional data, the distribution of synthetic data is represented by a probabilistic graphical model (e.g., a Bayesian network), while the raw data distribution is approximated by a collection of low-dimensional marginals.
Differential privacy (DP) is guaranteed by introducing random noise to each low-dimensional marginal distribution.
Despite its promising performance in practice, the statistical properties of marginal-based methods are rarely studied in the literature.
In this paper, we study DP data synthesis algorithms based on Bayesian networks (BN) from a statistical perspective. We establish a rigorous accuracy guarantee for BN-based algorithms, where the errors are measured by the total variation (TV) distance or the $L^2$ distance.
Related to downstream machine learning tasks, an upper bound for the utility error of the DP synthetic data is also derived. To complete the picture, we establish a lower bound for TV accuracy that holds for every $\epsilon$-DP synthetic data generator. | Accept: poster |
ICLR.cc/2019/Conference | Automatic generation of object shapes with desired functionalities | 3D objects (artefacts) are made to fulfill functions. Designing an object often starts with defining a list of functionalities that it should provide, also known as functional requirements. Today, the design of 3D object models is still a slow and largely artisanal activity, with few Computer-Aided Design (CAD) tools existing to aid the exploration of the design solution space. The purpose of the study is to explore the possibility of shape generation conditioned on desired functionalities. To accelerate the design process, we introduce an algorithm for generating object shapes with desired functionalities. We follow the principle form follows function, and assume that the form of a structure is correlated to its function. First, we use an artificial neural network to learn a function-to-form mapping by analysing a dataset of objects labeled with their functionalities. Then, we combine forms providing one or more desired functions, generating an object shape that is expected to provide all of them. Finally, we verify in simulation whether the generated object possesses the desired functionalities, by defining and executing functionality tests on it. | Reject |
ICLR.cc/2018/Conference | Beyond Finite Layer Neural Networks: Bridging Deep Architectures and Numerical Differential Equations | Deep neural networks have become the state-of-the-art models in numerous machine learning tasks. However, general guidance to network architecture design is still missing. In our work, we bridge deep neural network design with numerical differential equations. We show that many effective networks, such as ResNet, PolyNet, FractalNet and RevNet, can be interpreted as different numerical discretizations of differential equations. This finding brings us a brand new perspective on the design of effective deep architectures. We can take advantage of the rich knowledge in numerical analysis to guide us in designing new and potentially more effective deep networks. As an example, we propose a linear multi-step architecture (LM-architecture) which is inspired by the linear multi-step method solving ordinary differential equations. The LM-architecture is an effective structure that can be used on any ResNet-like networks. In particular, we demonstrate that LM-ResNet and LM-ResNeXt (i.e. the networks obtained by applying the LM-architecture on ResNet and ResNeXt respectively) can achieve noticeably higher accuracy than ResNet and ResNeXt on both CIFAR and ImageNet with comparable numbers of trainable parameters. In particular, on both CIFAR and ImageNet, LM-ResNet/LM-ResNeXt can significantly compress (>50%) the original networks while maintaining a similar performance. This can be explained mathematically using the concept of modified equation from numerical analysis. Last but not least, we also establish a connection between stochastic control and noise injection in the training process which helps to improve generalization of the networks. Furthermore, by relating stochastic training strategy with stochastic dynamic system, we can easily apply stochastic training to the networks with the LM-architecture. As an example, we introduced stochastic depth to LM-ResNet and achieve significant improvement over the original LM-ResNet on CIFAR10. | Invite to Workshop Track |
ICLR.cc/2021/Conference | Empirical Studies on the Convergence of Feature Spaces in Deep Learning | While deep learning is effective to learn features/representations from data, the distributions of samples in feature spaces learned by various architectures for different training tasks (e.g., latent layers of AEs and feature vectors in CNN classifiers) have not been well-studied or compared. We hypothesize that the feature spaces of networks trained by various architectures (AEs or CNNs) and tasks (supervised, unsupervised, or self-supervised learning) share some common subspaces, no matter what types of DNN architectures or whether the labels have been used in feature learning. To test our hypothesis, through Singular Value Decomposition (SVD) of feature vectors, we demonstrate that one could linearly project the feature vectors of the same group of samples to a similar distribution, where the distribution is represented as the top left singular vector (i.e., principal subspace of feature vectors), namely $\mathcal{P}$-vectors. We further assess the convergence of feature space learning using angles between $\mathcal{P}$-vectors obtained from the well-trained model and its checkpoint per epoch during the learning procedure, where a quasi-monotonic trend of convergence to small angles has been observed. Finally, we carry out case studies to connect $\mathcal{P}$-vectors to the data distribution, and generalization performance. Extensive experiments with practically-used MLP, AE and CNN architectures for classification, image reconstruction, and self-supervised learning tasks on MNIST, CIFAR-10 and CIFAR-100 datasets have been done to support our claims with solid evidences. | Reject |
ICLR.cc/2022/Conference | The Close Relationship Between Contrastive Learning and Meta-Learning | Contrastive learning has recently taken off as a paradigm for learning from unlabeled data. In this paper, we discuss the close relationship between contrastive learning and meta-learning under a certain task distribution. We complement this observation by showing that established meta-learning methods, such as Prototypical Networks, achieve comparable performance to SimCLR when paired with this task distribution. This relationship can be leveraged by taking established techniques from meta-learning, such as task-based data augmentation, and showing that they benefit contrastive learning as well. These tricks also benefit state-of-the-art self-supervised learners without using negative pairs such as BYOL, which achieves 94.6\% accuracy on CIFAR-10 using a self-supervised ResNet-18 feature extractor trained with our meta-learning tricks. We conclude that existing advances designed for contrastive learning or meta-learning can be exploited to benefit the other, and it is better for contrastive learning researchers to take lessons from the meta-learning literature (and vice-versa) than to reinvent the wheel. | Accept (Poster) |
ICLR.cc/2020/Conference | Generative Restricted Kernel Machines | We introduce a novel framework for generative models based on Restricted Kernel Machines (RKMs) with multi-view generation and uncorrelated feature learning capabilities, called Gen-RKM. To incorporate multi-view generation, this mechanism uses a shared representation of data from various views. The mechanism is flexible to incorporate both kernel-based, (deep) neural network and convolutional based models within the same setting. To update the parameters of the network, we propose a novel training procedure which jointly learns the features and shared representation. Experiments demonstrate the potential of the framework through qualitative evaluation of generated samples. | Reject |
ICLR.cc/2023/Conference | Understanding Hindsight Goal Relabeling Requires Rethinking Divergence Minimization | Hindsight goal relabeling has become a foundational technique for multi-goal reinforcement learning (RL). The idea is quite simple: any arbitrary trajectory can be seen as an expert demonstration for reaching the trajectory's end state. Intuitively, this procedure trains a goal-conditioned policy to imitate a sub-optimal expert. However, this connection between imitation and hindsight relabeling is not well understood. Modern imitation learning algorithms are described in the language of divergence minimization, and yet it remains an open problem how to recast hindsight goal relabeling into that framework. In this work, we develop a unified objective for goal-reaching that explains such a connection, from which we can derive goal-conditioned supervised learning (GCSL) and the reward function in hindsight experience replay (HER) from first principles. Experimentally, we find that despite recent advances in goal-conditioned behaviour cloning (BC), multi-goal Q-learning can still outperform BC-like methods; moreover, a vanilla combination of both actually hurts model performance. Under our framework, we study when BC is expected to help, and empirically validate our findings. Our work further bridges goal-reaching and generative modeling, illustrating the nuances and new pathways of extending the success of generative models to RL. | Reject |
ICLR.cc/2022/Conference | Learning the Representation of Behavior Styles with Imitation Learning | Imitation learning is one of the methods for reproducing expert demonstrations adaptively by learning a mapping between observations and actions. However, behavior styles such as motion trajectory and driving habit depend largely on the dataset of human maneuvers, and settle down to an average behavior style in most imitation learning algorithms. In this study, we propose a method named style behavior cloning (Style BC), which can not only infer the latent representation of behavior styles automatically, but also imitate different style policies from expert demonstrations. Our method is inspired by the word2vec algorithm and we construct a behavior-style to action mapping which is similar to the word-embedding to context mapping in word2vec. Empirical results on popular benchmark environments show that Style BC outperforms standard behavior cloning in prediction accuracy and expected reward significantly. Furthermore, compared with various baselines, our policy influenced by its assigned style embedding can better reproduce the expert behavior styles, especially in the complex environments or the number of the behavior styles is large. | Reject |
ICLR.cc/2023/Conference | Fuzzy Alignments in Directed Acyclic Graph for Non-Autoregressive Machine Translation | Non-autoregressive translation (NAT) reduces the decoding latency but suffers from performance degradation due to the multi-modality problem. Recently, the structure of directed acyclic graph has achieved great success in NAT, which tackles the multi-modality problem by introducing dependency between vertices. However, training it with negative log-likelihood loss implicitly requires a strict alignment between reference tokens and vertices, weakening its ability to handle multiple translation modalities. In this paper, we hold the view that all paths in the graph are fuzzily aligned with the reference sentence. We do not require the exact alignment but train the model to maximize a fuzzy alignment score between the graph and reference, which takes captured translations in all modalities into account. Extensive experiments on major WMT benchmarks show that our method substantially improves translation performance and increases prediction confidence, setting a new state of the art for NAT on the raw training data. | Accept: poster |
ICLR.cc/2022/Conference | Incorporating User-Item Similarity in Hybrid Neighborhood-based Recommendation System | Modern hybrid recommendation systems require a sufficient amount of data. However, several internet privacy issues make users skeptical about sharing their personal information with online service providers. This work introduces various novel methods utilizing the baseline estimate to learn user interests from their interactions. Subsequently, extracted user feature vectors are implemented to estimate the user-item correlations, providing an additional fine-tuning factor for neighborhood-based collaborative filtering systems. Comprehensive experiments show that utilizing the user-item similarity can boost the accuracy of hybrid neighborhood-based systems by at least $2.11\%$ while minimizing the need for tracking users' digital footprints. | Withdrawn |
ICLR.cc/2022/Conference | No Parameters Left Behind: Sensitivity Guided Adaptive Learning Rate for Training Large Transformer Models | Recent research has shown the existence of significant redundancy in large Transformer models. One can prune the redundant parameters without significantly sacrificing the generalization performance. However, we question whether the redundant parameters could have contributed more if they were properly trained. To answer this question, we propose a novel training strategy that encourages all parameters to be trained sufficiently. Specifically, we adaptively adjust the learning rate for each parameter according to its sensitivity, a robust gradient-based measure reflecting this parameter's contribution to the model performance. A parameter with low sensitivity is redundant, and we improve its fitting by increasing its learning rate. In contrast, a parameter with high sensitivity is well-trained, and we regularize it by decreasing its learning rate to prevent further overfitting. We conduct extensive experiments on natural language understanding, neural machine translation, and image classification to demonstrate the effectiveness of the proposed schedule. Analysis shows that the proposed schedule indeed reduces the redundancy and improves generalization performance. | Accept (Poster) |
ICLR.cc/2023/Conference | Self-Paced Learning Enhanced Physics-informed Neural Networks for Solving Partial Differential Equations | There is a hit discussion on solving partial differential equation by neural network. The famous PINN (physics-informed neural networks) has drawn worldwide attention since it was put forward. Despite its success in solving nonlinear partial differential equation, the difficulty in converging and the inefficiency in training process are definitely huge concerns. Normally, data for PINN is randomly chosen for a given distribution. Additionally, it's fitted to a model in a meaningless way. Curriculum Learning is a learning strategy that trains a model from easy samples to hard ones, which represents the meaningful human learning order. Self-paced Learning (SPL) is one of the significant branches of Automatic Curriculum Learning, which takes example-wise the training loss as Difficulty Measurer. SPL is an efficient strategy in enhancing the convergence rate of numerous models. In this paper, we propose a novel SPL-PINN learning framework, with SPL to accelerate the convergence progress of PINN. We demonstrate the effectiveness of SPL-PINN in a typical parabolic equation and Burgers equation. | Reject |
ICLR.cc/2022/Conference | Policy improvement by planning with Gumbel | AlphaZero is a powerful reinforcement learning algorithm based on approximate policy iteration and tree search. However, AlphaZero can fail to improve its policy network, if not visiting all actions at the root of a search tree. To address this issue, we propose a policy improvement algorithm based on sampling actions without replacement. Furthermore, we use the idea of policy improvement to replace the more heuristic mechanisms by which AlphaZero selects and uses actions, both at root nodes and at non-root nodes. Our new algorithms, Gumbel AlphaZero and Gumbel MuZero, respectively without and with model-learning, match the state of the art on Go, chess, and Atari, and significantly improve prior performance when planning with few simulations. | Accept (Spotlight) |
ICLR.cc/2021/Conference | Improved knowledge distillation by utilizing backward pass knowledge in neural networks | Knowledge distillation (KD) is one of the prominent techniques for model compression. In this method, the knowledge
of a large network (teacher) is distilled into a model (student) with usually significantly fewer parameters. KD tries to
better-match the output of the student model to that of the teacher model based on the knowledge extracts from
the forward pass of the teacher network. Although conventional KD is effective for matching the two networks over the given
data points, there is no guarantee that these models would match in other areas for which we do not have enough
training samples. In this work, we address that problem by generating new auxiliary training samples based on
extracting knowledge from the backward pass of the teacher in the areas where the student diverges greatly from the
teacher. We compute the difference between the teacher and the student and generate new data samples that
maximize the divergence. This is done by perturbing data samples in the direction of the gradient of the difference
between the student and the teacher. Augmenting the training set by adding this auxiliary improves the performance of
KD significantly and leads to a closer match between the student and the teacher. Using this approach, when data samples
come from a discrete domain, such as applications of natural language processing (NLP) and language understanding,
is not trivial. However, we show how this technique can be used successfully in such applications. We studied the effect
of the proposed method on various tasks in different domains, including images and NLP tasks with considerably
smaller student networks. The results of our experiments, when compared with the original KD, show 4% improvement on MNIST with a student network that is 160 times smaller, 1% improvement on a CIFAR-10 dataset with
a student that is 9 times smaller, and an average 1.5% improvement on the GLUE benchmark with a distilroBERTa-base
student. | Withdrawn |
ICLR.cc/2019/Conference | Transformer-XL: Language Modeling with Longer-Term Dependency | We propose a novel neural architecture, Transformer-XL, for modeling longer-term dependency. To address the limitation of fixed-length contexts, we introduce a notion of recurrence by reusing the representations from the history. Empirically, we show state-of-the-art (SoTA) results on both word-level and character-level language modeling datasets, including WikiText-103, One Billion Word, Penn Treebank, and enwiki8. Notably, we improve the SoTA results from 1.06 to 0.99 in bpc on enwiki8, from 33.0 to 18.9 in perplexity on WikiText-103, and from 28.0 to 23.5 in perplexity on One Billion Word. Performance improves when the attention length increases during evaluation, and our best model attends to up to 1,600 words and 3,800 characters. To quantify the effective length of dependency, we devise a new metric and show that on WikiText-103 Transformer-XL manages to model dependency that is about 80% longer than recurrent networks and 450% longer than Transformer. Moreover, Transformer-XL is up to 1,800+ times faster than vanilla Transformer during evaluation. | Reject |
ICLR.cc/2022/Conference | Automatic Concept Extraction for Concept Bottleneck-based Video Classification | Recent efforts in interpretable deep learning models have shown that concept-based explanation methods achieve competitive accuracy with standard end-to-end models and enable reasoning and intervention about extracted high-level visual concepts from images, e.g., identifying the wing color and beak length for bird-species classification. However, these concept bottleneck models rely on a domain expert providing a necessary and sufficient set of concepts--which is intractable for complex tasks such as video classification. For complex tasks, the labels and the relationship between visual elements span many frames, e.g., identifying a bird flying or catching prey--necessitating concepts with various levels of abstraction. To this end, we present CoDEx, an automatic Concept Discovery and Extraction module that rigorously composes a necessary and sufficient set of concept abstractions for concept-based video classification. CoDEx identifies a rich set of complex concept abstractions from natural language explanations of videos--obviating the need to predefine the amorphous set of concepts. To demonstrate our method’s viability, we construct two new public datasets that combine existing complex video classification datasets with short, crowd-sourced natural language explanations for their labels. Our method elicits inherent complex concept abstractions in natural language to generalize concept-bottleneck methods to complex tasks. | Reject |
ICLR.cc/2021/Conference | CNN Based Analysis of the Luria’s Alternating Series Test for Parkinson’s Disease Diagnostics | Deep-learning based image classification is applied in this studies to the Luria's alternating series tests to support diagnostics of the Parkinson's disease. Luria's alternating series tests belong to the family of fine-motor drawing tests and been used in neurology and psychiatry for nearly a century. Introduction of the digital tables and later tablet PCs has allowed deviating from the classical paper and pen setting, and observe kinematic and pressure parameters describing the test. While such setting has led to a highly accurate machine learning models, the visual component of the tests is left unused. Namely, the shapes of the drawn lines are not used to classify the drawings, which eventually has caused the shift in the assessment paradigm from visual-based to the numeric parameters based. The approach proposed in this paper allows combining two assessment paradigms by augmenting initial drawings by the kinematic and pressure parameters. The paper demonstrates that the resulting network has the accuracy similar to those of human practitioner. | Withdrawn |
ICLR.cc/2022/Conference | A Neural Tangent Kernel Perspective of Infinite Tree Ensembles | In practical situations, the tree ensemble is one of the most popular models along with neural networks. A soft tree is a variant of a decision tree. Instead of using a greedy method for searching splitting rules, the soft tree is trained using a gradient method in which the entire splitting operation is formulated in a differentiable form. Although ensembles of such soft trees have been used increasingly in recent years, little theoretical work has been done to understand their behavior. By considering an ensemble of infinite soft trees, this paper introduces and studies the Tree Neural Tangent Kernel (TNTK), which provides new insights into the behavior of the infinite ensemble of soft trees. Using the TNTK, we theoretically identify several non-trivial properties, such as global convergence of the training, the equivalence of the oblivious tree structure, and the degeneracy of the TNTK induced by the deepening of the trees. | Accept (Poster) |
ICLR.cc/2023/Conference | Revisiting Robustness in Graph Machine Learning | Many works show that node-level predictions of Graph Neural Networks (GNNs) are unrobust to small, often termed adversarial, changes to the graph structure. However, because manual inspection of a graph is difficult, it is unclear if the studied perturbations always preserve a core assumption of adversarial examples: that of unchanged semantic content. To address this problem, we introduce a more principled notion of an adversarial graph, which is aware of semantic content change. Using Contextual Stochastic Block Models (CSBMs) and real-world graphs, our results suggest: $i)$ for a majority of nodes the prevalent perturbation models include a large fraction of perturbed graphs violating the unchanged semantics assumption; $ii)$ surprisingly, all assessed GNNs show over-robustness - that is robustness beyond the point of semantic change. We find this to be a complementary phenomenon to adversarial examples and show that including the label-structure of the training graph into the inference process of GNNs significantly reduces over-robustness, while having a positive effect on test accuracy and adversarial robustness. Theoretically, leveraging our new semantics-aware notion of robustness, we prove that there is no robustness-accuracy tradeoff for inductively classifying a newly added node. | Accept: poster |
ICLR.cc/2020/Conference | VILD: Variational Imitation Learning with Diverse-quality Demonstrations | The goal of imitation learning (IL) is to learn a good policy from high-quality demonstrations. However, the quality of demonstrations in reality can be diverse, since it is easier and cheaper to collect demonstrations from a mix of experts and amateurs. IL in such situations can be challenging, especially when the level of demonstrators' expertise is unknown. We propose a new IL paradigm called Variational Imitation Learning with Diverse-quality demonstrations (VILD), where we explicitly model the level of demonstrators' expertise with a probabilistic graphical model and estimate it along with a reward function. We show that a naive estimation approach is not suitable to large state and action spaces, and fix this issue by using a variational approach that can be easily implemented using existing reinforcement learning methods. Experiments on continuous-control benchmarks demonstrate that VILD outperforms state-of-the-art methods. Our work enables scalable and data-efficient IL under more realistic settings than before. | Reject |
ICLR.cc/2018/Conference | Compact Encoding of Words for Efficient Character-level Convolutional Neural Networks Text Classification | This paper puts forward a new text to tensor representation that relies on information compression techniques to assign shorter codes to the most frequently used characters. This representation is language-independent with no need of pretraining and produces an encoding with no information loss. It provides an adequate description of the morphology of text, as it is able to represent prefixes, declensions, and inflections with similar vectors and are able to represent even unseen words on the training dataset. Similarly, as it is compact yet sparse, is ideal for speed up training times using tensor processing libraries. As part of this paper, we show that this technique is especially effective when coupled with convolutional neural networks (CNNs) for text classification at character-level. We apply two variants of CNN coupled with it. Experimental results show that it drastically reduces the number of parameters to be optimized, resulting in competitive classification accuracy values in only a fraction of the time spent by one-hot encoding representations, thus enabling training in commodity hardware. | Reject |
ICLR.cc/2020/Conference | Model-based Saliency for the Detection of Adversarial Examples | Adversarial perturbations cause a shift in the salient features of an image, which may result in a misclassification. We demonstrate that gradient-based saliency approaches are unable to capture this shift, and develop a new defense which detects adversarial examples based on learnt saliency models instead. We study two approaches: a CNN trained to distinguish between natural and adversarial images using the saliency masks produced by our learnt saliency model, and a CNN trained on the salient pixels themselves as its input. On MNIST, CIFAR-10 and ASSIRA, our defenses are able to detect various adversarial attacks, including strong attacks such as C&W and DeepFool, contrary to gradient-based saliency and detectors which rely on the input image. The latter are unable to detect adversarial images when the L_2- and L_infinity- norms of the perturbations are too small. Lastly, we find that the salient pixel based detector improves on saliency map based detectors as it is more robust to white-box attacks. | Reject |
ICLR.cc/2021/Conference | Signal Coding and Reconstruction using Spike Trains | In many animal sensory pathways, the transformation from external stimuli to spike trains is essentially deterministic. In this context, a new mathematical framework for coding and reconstruction, based on a biologically plausible model of the spiking neuron, is presented. The framework considers encoding of a signal through spike trains generated by an ensemble of neurons via a standard convolve-then-threshold mechanism, albeit with a wide variety of convolution kernels. Neurons are distinguished by their convolution kernels and threshold values. Reconstruction is posited as a convex optimization minimizing energy. Formal conditions under which perfect reconstruction of the signal from the spike trains is possible are then identified. Coding experiments on a large audio dataset are presented to demonstrate the strength of the framework. | Reject |
ICLR.cc/2021/Conference | Using MMD GANs to correct physics models and improve Bayesian parameter estimation | Bayesian parameter estimation methods are robust techniques for quantifying properties of physical systems which cannot be observed directly. In estimating such parameters, one first requires a physics model of the phenomenon to be studied. Often, such a model follows a series of assumptions to make parameter inference feasible. When simplified models are used for inference, however, systematic differences between model predictions and observed data may propagate throughout the parameter estimation process, biasing inference results. In this work, we use generative adversarial networks (GANs) based on the maximum mean discrepancy (MMD) to learn small stochastic corrections to physics models in order to minimize inference bias. We further propose a hybrid training procedure utilizing both the MMD and the standard GAN objective functionals. We demonstrate the ability to learn stochastic model corrections and eliminate inference bias on a toy problem wherein the true data distribution is known. Subsequently, we apply these methods to a mildly ill-posed inference problem in magnetic resonance imaging (MRI), showing improvement over an established inference method. Finally, because 3D MRI images often contain millions of voxels which would each require parameter inference, we train a conditional variational autoencoder (CVAE) network on the corrected MRI physics model to perform fast inference and make this approach practical.
| Withdrawn |
ICLR.cc/2021/Conference | Fidelity-based Deep Adiabatic Scheduling | Adiabatic quantum computation is a form of computation that acts by slowly interpolating a quantum system between an easy to prepare initial state and a final state that represents a solution to a given computational problem. The choice of the interpolation schedule is critical to the performance: if at a certain time point, the evolution is too rapid, the system has a high probability to transfer to a higher energy state, which does not represent a solution to the problem. On the other hand, an evolution that is too slow leads to a loss of computation time and increases the probability of failure due to decoherence. In this work, we train deep neural models to produce optimal schedules that are conditioned on the problem at hand. We consider two types of problem representation: the Hamiltonian form, and the Quadratic Unconstrained Binary Optimization (QUBO) form. A novel loss function that scores schedules according to their approximated success probability is introduced. We benchmark our approach on random QUBO problems, Grover search, 3-SAT, and MAX-CUT problems and show that our approach outperforms, by a sizable margin, the linear schedules as well as alternative approaches that were very recently proposed. | Accept (Spotlight) |
ICLR.cc/2019/Conference | Context Mover's Distance & Barycenters: Optimal transport of contexts for building representations | We propose a unified framework for building unsupervised representations of entities and their compositions, by viewing each entity as a histogram (or distribution) over its contexts. This enables us to take advantage of optimal transport and construct representations that effectively harness the geometry of the underlying space containing the contexts. Our method captures uncertainty via modelling the entities as distributions and simultaneously provides interpretability with the optimal transport map, hence giving a novel perspective for building rich and powerful feature representations. As a guiding example, we formulate unsupervised representations for text, and demonstrate it on tasks such as sentence similarity and word entailment detection. Empirical results show strong advantages gained through the proposed framework. This approach can potentially be used for any unsupervised or supervised problem (on text or other modalities) with a co-occurrence structure, such as any sequence data. The key tools at the core of this framework are Wasserstein distances and Wasserstein barycenters. | Reject |
ICLR.cc/2023/Conference | Multi-Level Contrastive Learning for Dense Prediction Task | In this work, we present Multi-Level Contrastive Learning for Dense Prediction Task (MCL), an efficient self-supervised method to learn region-level feature representation for dense prediction tasks. This approach is motivated by the three key factors in detection: localization, scale consistency and recognition. Considering the above factors, we design a novel pretext task, which explicitly encodes absolute position and scale information simultaneously by assembling multi-scale images in a montage manner to mimic multi-object scenario. Unlike the existing image-level self-supervised methods, our method constructs a multi-level contrastive loss by considering each sub-region of the montage image as a singleton to learn a regional semantic representation for translation and scale consistency, while reducing the pre-training epochs to the same as supervised pre-training. Extensive experiments show that MCL consistently outperforms the recent state-of-the-art methods on various datasets with significant margins. In particular, MCL obtains 42.5 AP^bb and 38.3 AP^mk on COCO with the 1x schedule and surpasses MoCo by 4.0 AP^bb and 3.1 AP^mk, when using Mask R-CNN with an R50-FPN backbone pre-trained with 100 epochs. In addition, we further explore the alignment between pretext task and downstream tasks. We extend our pretext task to supervised pre-training, which achieves a similar performance with self-supervised learning, demonstrating the importance of the alignment between pretext task and downstream tasks. | Withdrawn |
ICLR.cc/2018/Conference | HyperNetworks with statistical filtering for defending adversarial examples | Deep learning algorithms have been known to be vulnerable to adversarial perturbations in various tasks such as image classification. This problem was addressed by employing several defense methods for detection and rejection of particular types of attacks. However, training and manipulating networks according to particular defense schemes increases computational complexity of the learning algorithms. In this work, we propose a simple yet effective method to improve robustness of convolutional neural networks (CNNs) to adversarial attacks by using data dependent adaptive convolution kernels. To this end, we propose a new type of HyperNetwork in order to employ statistical properties of input data and features for computation of statistical adaptive maps. Then, we filter convolution weights of CNNs with the learned statistical maps to compute dynamic kernels. Thereby, weights and kernels are collectively optimized for learning of image classification models robust to
adversarial attacks without employment of additional target detection and rejection algorithms.
We empirically demonstrate that the proposed method enables CNNs to spontaneously defend against different types of attacks, e.g. attacks generated by Gaussian noise, fast gradient sign methods (Goodfellow et al., 2014) and a black-box attack (Narodytska & Kasiviswanathan, 2016). | Withdrawn |
ICLR.cc/2022/Conference | Learning to Extend Molecular Scaffolds with Structural Motifs | Recent advancements in deep learning-based modeling of molecules promise to accelerate in silico drug discovery. A plethora of generative models is available, building molecules either atom-by-atom and bond-by-bond or fragment-by-fragment. However, many drug discovery projects require a fixed scaffold to be present in the generated molecule, and incorporating that constraint has only recently been explored. Here, we propose MoLeR, a graph-based model that naturally supports scaffolds as initial seed of the generative procedure, which is possible because it is not conditioned on the generation history. Our experiments show that MoLeR performs comparably to state-of-the-art methods on unconstrained molecular optimization tasks, and outperforms them on scaffold-based tasks, while being an order of magnitude faster to train and sample from than existing approaches. Furthermore, we show the influence of a number of seemingly minor design choices on the overall performance. | Accept (Poster) |
ICLR.cc/2023/Conference | Generalizable Multi-Relational Graph Representation Learning: A Message Intervention Approach | With the edges associated with labels and directions, the so-called multi-relational graph possesses powerful expressiveness, which is beneficial to many applications. However, as the heterogeneity brought by the higher cardinality of edges and relations climbs up, more trivial relations are taken into account for the downstream task since they are often highly correlated to the target. As a result, with being forced to fit the non-causal relational patterns on the training set, the downstream model, like graph neural network (GNN), may suffer from poor generalizability on the testing set since the inference is mainly made according to misleading clues. In this paper, under the paradigm of graph convolution, we probe the multi-relational message passing process from the perspective of causality and then propose a Message Intervention method for learning generalizable muLtirElational gRaph representations, coined MILER. In particular, MILER first encodes the vertices and relations into embeddings with relational and directional awareness, then a message diverter is employed to split the original message flow into two flows of interest, i.e., the causal and trivial message flows. Afterward, the message intervention is carried out with the guidance of the backdoor adjustment rule. Extensive experiments on several knowledge graph benchmarks validate the effectiveness as well as the superior generalization ability of MILER. | Withdrawn |
ICLR.cc/2021/Conference | Learning advanced mathematical computations from examples | Using transformers over large generated datasets, we train models to learn mathematical properties of differential systems, such as local stability, behavior at infinity and controllability. We achieve near perfect prediction of qualitative characteristics, and good approximations of numerical features of the system. This demonstrates that neural networks can learn to perform complex computations, grounded in advanced theory, from examples, without built-in mathematical knowledge. | Accept (Poster) |
ICLR.cc/2021/Conference | Bayesian Online Meta-Learning | Neural networks are known to suffer from catastrophic forgetting when trained on sequential datasets. While there have been numerous attempts to solve this problem for large-scale supervised classification, little has been done to overcome catastrophic forgetting for few-shot classification problems. Few-shot meta-learning algorithms often require all few-shot tasks to be readily available in a batch for training. The popular gradient-based model-agnostic meta-learning algorithm (MAML) is a typical algorithm that suffers from these limitations. This work introduces a Bayesian online meta-learning framework to tackle the catastrophic forgetting and the sequential few-shot tasks problems. Our framework incorporates MAML into a Bayesian online learning algorithm with Laplace approximation or variational inference. This framework enables few-shot classification on a range of sequentially arriving datasets with a single meta-learned model and training on sequentially arriving few-shot tasks. The experimental evaluations demonstrate that our framework can effectively prevent catastrophic forgetting and is capable of online meta-learning in various few-shot classification settings. | Reject |
ICLR.cc/2022/Conference | Towards Structured Dynamic Sparse Pre-Training of BERT | Identifying algorithms for computational efficient unsupervised training of large language models is an important and active area of research.
In this work, we develop and study a straightforward, dynamic always-sparse pre-training approach for BERT language modeling, which leverages periodic compression steps based on magnitude pruning followed by random parameter re-allocation.
This approach enables us to achieve Pareto improvements in terms of the number of floating-point operations (FLOPs) over statically sparse and dense models across a broad spectrum of network sizes.
Furthermore, we demonstrate that training remains FLOP-efficient when using coarse-grained block sparsity, making it particularly promising for efficient execution on modern hardware accelerators. | Reject |
End of preview. Expand
in Dataset Viewer.
- Downloads last month
- 35