Paper Group NANR 15
Detecting and Diagnosing Adversarial Images with Class-Conditional Capsule Reconstructions. Axial Attention in Multidimensional Transformers. Learning Self-Correctable Policies and Value Functions from Demonstrations with Negative Sampling. Smooth Kernels Improve Adversarial Robustness and Perceptually-Aligned Gradients. Robust Subspace Recovery La …
Detecting and Diagnosing Adversarial Images with Class-Conditional Capsule Reconstructions
| Title | Detecting and Diagnosing Adversarial Images with Class-Conditional Capsule Reconstructions |
| Authors | Anonymous |
| Abstract | Adversarial examples raise questions about whether neural network models are sensitive to the same visual features as humans. In this paper, we first detect adversarial examples or otherwise corrupted images based on a class-conditional reconstruction of the input. To specifically attack our detection mechanism, we propose the Reconstructive Attack which seeks both to cause a misclassification and a low reconstruction error. This reconstructive attack produces undetected adversarial examples but with much smaller success rate. Among all these attacks, we find that CapsNets always perform better than convolutional networks. Then, we diagnose the adversarial examples for CapsNets and find that the success of the reconstructive attack was proportional to the visual similarity between the source and target class. Additionally, the resulting perturbations can cause the input image to appear visually more like the target class and hence become non-adversarial. These suggest that CapsNets use features that are more aligned with human perception and address the central issue raised by adversarial examples. |
| Tasks | |
| Published | 2020-01-01 |
| URL | https://openreview.net/forum?id=Skgy464Kvr |
| https://openreview.net/pdf?id=Skgy464Kvr | |
| PWC | https://paperswithcode.com/paper/detecting-and-diagnosing-adversarial-images-1 |
| Repo | |
| Framework | |
Axial Attention in Multidimensional Transformers
| Title | Axial Attention in Multidimensional Transformers |
| Authors | Anonymous |
| Abstract | Self-attention effectively captures large receptive fields with high information bandwidth, but its computational resource requirements grow quadratically with the number of points over which attention is performed. For data arranged as large multidimensional tensors, such as images and videos, the quadratic growth makes self-attention prohibitively expensive. These tensors often have thousands of positions that one wishes to capture and proposed attentional alternatives either limit the resulting receptive field or require custom subroutines. We propose Axial Attention, a simple generalization of self-attention that naturally aligns with the multiple dimensions of the tensors in both the encoding and the decoding settings. The Axial Transformer uses axial self-attention layers and a shift operation to efficiently build large and full receptive fields. Notably the proposed structure of the layers allows for the vast majority of the context to be computed in parallel during decoding without introducing any independence assumptions. This semi-parallel structure goes a long way to making decoding from even a very large Axial Transformer broadly applicable. We demonstrate state-of-the-art results for the Axial Transformer on the ImageNet-32 and ImageNet-64 image benchmarks as well as on the BAIR Robotic Pushing video benchmark. We open source the implementation of Axial Transformers. |
| Tasks | |
| Published | 2020-01-01 |
| URL | https://openreview.net/forum?id=H1e5GJBtDr |
| https://openreview.net/pdf?id=H1e5GJBtDr | |
| PWC | https://paperswithcode.com/paper/axial-attention-in-multidimensional |
| Repo | |
| Framework | |
Learning Self-Correctable Policies and Value Functions from Demonstrations with Negative Sampling
| Title | Learning Self-Correctable Policies and Value Functions from Demonstrations with Negative Sampling |
| Authors | Anonymous |
| Abstract | Imitation learning, followed by reinforcement learning algorithms, is a promising paradigm to solve complex control tasks sample-efficiently. However, learning from demonstrations often suffers from the covariate shift problem, which results in cascading errors of the learned policy. We introduce a notion of conservatively extrapolated value functions, which provably lead to policies with self-correction. We design an algorithm Value Iteration with Negative Sampling (VINS) that practically learns such value functions with conservative extrapolation. We show that VINS can correct mistakes of the behavioral cloning policy on simulated robotics benchmark tasks. We also propose the algorithm of using VINS to initialize a reinforcement learning algorithm, which is shown to outperform prior works in sample efficiency. |
| Tasks | Imitation Learning |
| Published | 2020-01-01 |
| URL | https://openreview.net/forum?id=rke-f6NKvS |
| https://openreview.net/pdf?id=rke-f6NKvS | |
| PWC | https://paperswithcode.com/paper/learning-self-correctable-policies-and-value-1 |
| Repo | |
| Framework | |
Smooth Kernels Improve Adversarial Robustness and Perceptually-Aligned Gradients
| Title | Smooth Kernels Improve Adversarial Robustness and Perceptually-Aligned Gradients |
| Authors | Haohan Wang, Xindi Wu, Songwei Ge, Zachary C. Lipton, Eric P. Xing |
| Abstract | Recent research has shown that CNNs are often overly sensitive to high-frequency textural patterns. Inspired by the intuition that humans are more sensitive to the lower-frequency (larger-scale) patterns we design a regularization scheme that penalizes large differences between adjacent components within each convolutional kernel. We apply our regularization onto several popular training methods, demonstrating that the models with the proposed smooth kernels enjoy improved adversarial robustness. Further, building on recent work establishing connections between adversarial robustness and interpretability, we show that our method appears to give more perceptually-aligned gradients. |
| Tasks | |
| Published | 2020-01-01 |
| URL | https://openreview.net/forum?id=BJerUCEtPB |
| https://openreview.net/pdf?id=BJerUCEtPB | |
| PWC | https://paperswithcode.com/paper/smooth-kernels-improve-adversarial-robustness |
| Repo | |
| Framework | |
Robust Subspace Recovery Layer for Unsupervised Anomaly Detection
| Title | Robust Subspace Recovery Layer for Unsupervised Anomaly Detection |
| Authors | Anonymous |
| Abstract | We propose a neural network for unsupervised anomaly detection with a novel robust subspace recovery layer (RSR layer). This layer seeks to extract the underlying subspace from a latent representation of the given data and removes outliers that lie away from this subspace. It is used within an autoencoder. The encoder maps the data into a latent space, from which the RSR layer extracts the subspace. The decoder then smoothly maps back the underlying subspace to a ``manifold” close to the original inliers. Inliers and outliers are distinguished according to the distances between the original and mapped positions (small for inliers and large for outliers). Extensive numerical experiments with both image and document datasets demonstrate state-of-the-art precision and recall. | |
| Tasks | Anomaly Detection, Unsupervised Anomaly Detection |
| Published | 2020-01-01 |
| URL | https://openreview.net/forum?id=rylb3eBtwr |
| https://openreview.net/pdf?id=rylb3eBtwr | |
| PWC | https://paperswithcode.com/paper/robust-subspace-recovery-layer-for-1 |
| Repo | |
| Framework | |
WEEGNET: an wavelet based Convnet for Brain-computer interfaces
| Title | WEEGNET: an wavelet based Convnet for Brain-computer interfaces |
| Authors | Anonymous |
| Abstract | Brain-computer interfaces (BCI) are systems that link the brain with machines using brainwaves as a medium of communication using electroencephalography to explore the brain activity which is an affordable solution, noninvasive, easy setup, and portability. However, the neural signals are noisy, non-stationary, and nonlinear where the processing of those signals in a pattern recognition problem needs a complex pipeline of preprocessing, feature extraction, and classification algorithms that need an apriori knowledge to avoid compatibility issues and a deep understanding of the studied signals. Moreover, some techniques need a huge computational power on the CPU and a huge size of RAM. Therefore, several papers proposed to use Deep Learning to get state of the art performance and visualization of the learned features to have more understanding about the neural signals. But, the convolutional neural network (Convnet) are not used properly and the results are often random when we reproduced the works. Hence, we propose a combination of the discrete wavelet transform (DWT) and a Convnet that processes raw EEG data. The DWT will be used to reduce the size of the data without losing useful information. Also, a modified version of EEGNET will be used to extract the features and classification. |
| Tasks | EEG |
| Published | 2020-01-01 |
| URL | https://openreview.net/forum?id=HJeN6grYDr |
| https://openreview.net/pdf?id=HJeN6grYDr | |
| PWC | https://paperswithcode.com/paper/weegnet-an-wavelet-based-convnet-for-brain |
| Repo | |
| Framework | |
Deep Reinforcement Learning with Implicit Human Feedback
| Title | Deep Reinforcement Learning with Implicit Human Feedback |
| Authors | Duo Xu, Mohit Agarwal, Raghupathy Sivakumar, Faramarz Fekri |
| Abstract | We consider the following central question in the field of Deep Reinforcement Learning (DRL): How can we use implicit human feedback to accelerate and optimize the training of a DRL algorithm? State-of-the-art methods rely on any human feedback to be provided explicitly, requiring the active participation of humans (e.g., expert labeling, demonstrations, etc.). In this work, we investigate an alternative paradigm, where non-expert humans are silently observing (and assessing) the agent interacting with the environment. The human’s intrinsic reactions to the agent’s behavior is sensed as implicit feedback by placing electrodes on the human scalp and monitoring what are known as event-related electric potentials. The implicit feedback is then used to augment the agent’s learning in the RL tasks. We develop a system to obtain and accurately decode the implicit human feedback (specifically error-related event potentials) for state-action pairs in an Atari-type environment. As a baseline contribution, we demonstrate the feasibility of capturing error-potentials of a human observer watching an agent learning to play several different Atari-games using an electroencephalogram (EEG) cap, and then decoding the signals appropriately and using them as an auxiliary reward function to a DRL algorithm with the intent of accelerating its learning of the game. Building atop the baseline, we then make the following novel contributions in our work: (i) We argue that the definition of error-potentials is generalizable across different environments; specifically we show that error-potentials of an observer can be learned for a specific game, and the definition used as-is for another game without requiring re-learning of the error-potentials. (ii) We propose two different frameworks to combine recent advances in DRL into the error-potential based feedback system in a sample-efficient manner, allowing humans to provide implicit feedback while training in the loop, or prior to the training of the RL agent. (iii) Finally, we scale the implicit human feedback (via ErrP) based RL to reasonably complex environments (games) and demonstrate the significance of our approach through synthetic and real user experiments. |
| Tasks | Atari Games, EEG |
| Published | 2020-01-01 |
| URL | https://openreview.net/forum?id=rJgDT04twH |
| https://openreview.net/pdf?id=rJgDT04twH | |
| PWC | https://paperswithcode.com/paper/deep-reinforcement-learning-with-implicit |
| Repo | |
| Framework | |
Conditional Flow Variational Autoencoders for Structured Sequence Prediction
| Title | Conditional Flow Variational Autoencoders for Structured Sequence Prediction |
| Authors | Anonymous |
| Abstract | Prediction of future states of the environment and interacting agents is a key competence required for autonomous agents to operate successfully in the real world. Prior work for structured sequence prediction based on latent variable models imposes a uni-modal standard Gaussian prior on the latent variables. This induces a strong model bias which makes it challenging to fully capture the multi-modality of the distribution of the future states. In this work, we introduce Conditional Flow Variational Autoencoders (CF-VAE) using our novel conditional normalizing flow based prior to capture complex multi-modal conditional distributions for effective structured sequence prediction. Moreover, we propose two novel regularization schemes which stabilizes training and deals with posterior collapse for stable training and better match to the data distribution. Our experiments on three multi-modal structured sequence prediction datasets – MNIST Sequences, Stanford Drone and HighD – show that the proposed method obtains state of art results across different evaluation metrics. |
| Tasks | Latent Variable Models |
| Published | 2020-01-01 |
| URL | https://openreview.net/forum?id=BklmtJBKDB |
| https://openreview.net/pdf?id=BklmtJBKDB | |
| PWC | https://paperswithcode.com/paper/conditional-flow-variational-autoencoders-for-1 |
| Repo | |
| Framework | |
ODE Analysis of Stochastic Gradient Methods with Optimism and Anchoring for Minimax Problems and GANs
| Title | ODE Analysis of Stochastic Gradient Methods with Optimism and Anchoring for Minimax Problems and GANs |
| Authors | Anonymous |
| Abstract | Despite remarkable empirical success, the training dynamics of generative adversarial networks (GAN), which involves solving a minimax game using stochastic gradients, is still poorly understood. In this work, we analyze last-iterate convergence of simultaneous gradient descent (simGD) and its variants under the assumption of convex-concavity, guided by a continuous-time analysis with differential equations. First, we show that simGD, as is, converges with stochastic sub-gradients under strict convexity in the primal variable. Second, we generalize optimistic simGD to accommodate an optimism rate separate from the learning rate and show its convergence with full gradients. Finally, we present anchored simGD, a new method, and show convergence with stochastic subgradients. |
| Tasks | |
| Published | 2020-01-01 |
| URL | https://openreview.net/forum?id=BygIjTNtPr |
| https://openreview.net/pdf?id=BygIjTNtPr | |
| PWC | https://paperswithcode.com/paper/ode-analysis-of-stochastic-gradient-methods-1 |
| Repo | |
| Framework | |
Optimistic Exploration even with a Pessimistic Initialisation
| Title | Optimistic Exploration even with a Pessimistic Initialisation |
| Authors | Anonymous |
| Abstract | Optimistic initialisation is an effective strategy for efficient exploration in reinforcement learning (RL). In the tabular case, all provably efficient model-free algorithms rely on it. However, model-free deep RL algorithms do not use optimistic initialisation despite taking inspiration from these provably efficient tabular algorithms. In particular, in scenarios with only positive rewards, Q-values are initialised at their lowest possible values due to commonly used network initialisation schemes, a pessimistic initialisation. Merely initialising the network to output optimistic Q-values is not enough, since we cannot ensure that they remain optimistic for novel state-action pairs, which is crucial for exploration. We propose a simple count-based augmentation to pessimistically initialised Q-values that separates the source of optimism from the neural network. We show that this scheme is provably efficient in the tabular setting and extend it to the deep RL setting. Our algorithm, Optimistic Pessimistically Initialised Q-Learning (OPIQ), augments the Q-value estimates of a DQN-based agent with count-derived bonuses to ensure optimism during both action selection and bootstrapping. We show that OPIQ outperforms non-optimistic DQN variants that utilise a pseudocount-based intrinsic motivation in hard exploration tasks, and that it predicts optimistic estimates for novel state-action pairs. |
| Tasks | Efficient Exploration, Q-Learning |
| Published | 2020-01-01 |
| URL | https://openreview.net/forum?id=r1xGP6VYwH |
| https://openreview.net/pdf?id=r1xGP6VYwH | |
| PWC | https://paperswithcode.com/paper/optimistic-exploration-even-with-a |
| Repo | |
| Framework | |
A Novel Analysis Framework of Lower Complexity Bounds for Finite-Sum Optimization
| Title | A Novel Analysis Framework of Lower Complexity Bounds for Finite-Sum Optimization |
| Authors | Anonymous |
| Abstract | 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. |
| Tasks | |
| Published | 2020-01-01 |
| URL | https://openreview.net/forum?id=SygD31HFvB |
| https://openreview.net/pdf?id=SygD31HFvB | |
| PWC | https://paperswithcode.com/paper/a-novel-analysis-framework-of-lower |
| Repo | |
| Framework | |
Swoosh! Rattle! Thump! - Actions that Sound
| Title | Swoosh! Rattle! Thump! - Actions that Sound |
| Authors | Anonymous |
| Abstract | Truly intelligent agents need to capture the interplay of all their senses to build a rich physical understanding of their world. In robotics, we have seen tremendous progress in using visual and tactile perception; however we have often ignored a key sense: sound. This is primarily due to lack of data that captures the interplay of action and sound. In this work, we perform the first large-scale study of the interactions between sound and robotic action. To do this, we create the largest available sound-action-vision dataset with 15,000 interactions on 60 objects using our robotic platform Tilt-Bot. By tilting objects and allowing them to crash into the walls of a robotic tray, we collect rich four-channel audio information. Using this data, we explore the synergies between sound and action, and present three key insights. First, sound is indicative of fine-grained object class information, e.g., sound can differentiate a metal screwdriver from a metal wrench. Second, sound also contains information about the causal effects of an action, i.e. given the sound produced, we can predict what action was applied on the object. Finally, object representations derived from audio embeddings are indicative of implicit physical properties. We demonstrate that on previously unseen objects, audio embeddings generated through interactions can predict forward models 24% better than passive visual embeddings. |
| Tasks | |
| Published | 2020-01-01 |
| URL | https://openreview.net/forum?id=SJeUm1HtDH |
| https://openreview.net/pdf?id=SJeUm1HtDH | |
| PWC | https://paperswithcode.com/paper/swoosh-rattle-thump-actions-that-sound |
| Repo | |
| Framework | |
Deep Reasoning Networks: Thinking Fast and Slow, for Pattern De-mixing
| Title | Deep Reasoning Networks: Thinking Fast and Slow, for Pattern De-mixing |
| Authors | Anonymous |
| Abstract | We introduce Deep Reasoning Networks (DRNets), an end-to-end framework that combines deep learning with reasoning for solving pattern de-mixing problems, typically in an unsupervised or weakly-supervised setting. DRNets exploit problem structure and prior knowledge by tightly combining logic and constraint reasoning with stochastic-gradient-based neural network optimization. We illustrate the power of DRNets on de-mixing overlapping hand-written Sudokus (Multi-MNIST-Sudoku) and on a substantially more complex task in scientific discovery that concerns inferring crystal structures of materials from X-ray diffraction data (Crystal-Structure-Phase-Mapping). DRNets significantly outperform the state of the art and experts’ capabilities on Crystal-Structure-Phase-Mapping, recovering more precise and physically meaningful crystal structures. On Multi-MNIST-Sudoku, DRNets perfectly recovered the mixed Sudokus’ digits, with 100% digit accuracy, outperforming the supervised state-of-the-art MNIST de-mixing models. |
| Tasks | |
| Published | 2020-01-01 |
| URL | https://openreview.net/forum?id=HkehD3VtvS |
| https://openreview.net/pdf?id=HkehD3VtvS | |
| PWC | https://paperswithcode.com/paper/deep-reasoning-networks-thinking-fast-and |
| Repo | |
| Framework | |
Robust Federated Learning Through Representation Matching and Adaptive Hyper-parameters
| Title | Robust Federated Learning Through Representation Matching and Adaptive Hyper-parameters |
| Authors | Anonymous |
| Abstract | Federated learning is a distributed, privacy-aware learning scenario which trains a single model on data belonging to several clients. Each client trains a local model on its data and the local models are then aggregated by a central party. Current federated learning methods struggle in cases with heterogeneous client-side data distributions which can quickly lead to divergent local models and a collapse in performance. Careful hyper-parameter tuning is particularly important in these cases but traditional automated hyper-parameter tuning methods would require several training trials which is often impractical in a federated learning setting. We describe a two-pronged solution to the issues of robustness and hyper-parameter tuning in federated learning settings. We propose a novel representation matching scheme that reduces the divergence of local models by ensuring the feature representations in the global (aggregate) model can be derived from the locally learned representations. We also propose an online hyper-parameter tuning scheme which uses an online version of the REINFORCE algorithm to find a hyper-parameter distribution that maximizes the expected improvements in training loss. We show on several benchmarks that our two-part scheme of local representation matching and global adaptive hyper-parameters significantly improves performance and training robustness. |
| Tasks | |
| Published | 2020-01-01 |
| URL | https://openreview.net/forum?id=SJeeL04KvH |
| https://openreview.net/pdf?id=SJeeL04KvH | |
| PWC | https://paperswithcode.com/paper/robust-federated-learning-through |
| Repo | |
| Framework | |
MONET: Debiasing Graph Embeddings via the Metadata-Orthogonal Training Unit
| Title | MONET: Debiasing Graph Embeddings via the Metadata-Orthogonal Training Unit |
| Authors | Anonymous |
| Abstract | Are Graph Neural Networks (GNNs) fair? In many real world graphs, the formation of edges is related to certain node attributes (e.g. gender, community, reputation). In this case, any GNN using these edges will be biased by this information, as it is encoded in the structure of the adjacency matrix itself. In this paper, we show that when metadata is correlated with the formation of node neighborhoods, unsupervised node embedding dimensions learn this metadata. This bias implies an inability to control for important covariates in real-world applications, such as recommendation systems. To solve these issues, we introduce the Metadata-Orthogonal Node Embedding Training (MONET) unit, a general model for debiasing embeddings of nodes in a graph. MONET achieves this by ensuring that the node embeddings are trained on a hyperplane orthogonal to that of the node metadata. This effectively organizes unstructured embedding dimensions into an interpretable topology-only, metadata-only division with no linear interactions. We illustrate the effectiveness of MONET though our experiments on a variety of real world graphs, which shows that our method can learn and remove the effect of arbitrary covariates in tasks such as preventing the leakage of political party affiliation in a blog network, and thwarting the gaming of embedding-based recommendation systems. |
| Tasks | Recommendation Systems |
| Published | 2020-01-01 |
| URL | https://openreview.net/forum?id=rkx3-04FwB |
| https://openreview.net/pdf?id=rkx3-04FwB | |
| PWC | https://paperswithcode.com/paper/monet-debiasing-graph-embeddings-via-the-1 |
| Repo | |
| Framework | |
