Paper Group NANR 61
Improving Evolutionary Strategies with Generative Neural Networks. Adaptive Learned Bloom Filter (Ada-BF): Efficient Utilization of the Classifier. Disagreement-Regularized Imitation Learning. Learning Generative Image Object Manipulations from Language Instructions. Power up! Robust Graph Convolutional Network based on Graph Powering. Compressive …
Improving Evolutionary Strategies with Generative Neural Networks
| Title | Improving Evolutionary Strategies with Generative Neural Networks |
| Authors | Anonymous |
| Abstract | Evolutionary Strategies (ES) are a popular family of black-box zeroth-order optimization algorithms which rely on search distributions to efficiently optimize a large variety of objective functions. This paper investigates the potential benefits of using highly flexible search distributions in ES algorithms, in contrast to standard ones (typically Gaussians). We model such distributions with Generative Neural Networks (GNNs) and introduce a new ES algorithm that leverages their expressiveness to accelerate the stochastic search. Because it acts as a plug-in, our approach allows to augment virtually any standard ES algorithm with flexible search distributions. We demonstrate the empirical advantages of this method on a diversity of objective functions. |
| Tasks | |
| Published | 2020-01-01 |
| URL | https://openreview.net/forum?id=SJlDDnVKwS |
| https://openreview.net/pdf?id=SJlDDnVKwS | |
| PWC | https://paperswithcode.com/paper/improving-evolutionary-strategies-with-1 |
| Repo | |
| Framework | |
Adaptive Learned Bloom Filter (Ada-BF): Efficient Utilization of the Classifier
| Title | Adaptive Learned Bloom Filter (Ada-BF): Efficient Utilization of the Classifier |
| Authors | Anonymous |
| Abstract | Recent work suggests improving the performance of Bloom filter by incorporating a machine learning model as a binary classifier. However, such learned Bloom filter does not take full advantage of the predicted probability scores. We proposed new algorithms that generalize the learned Bloom filter by using the complete spectrum of the scores regions. We proved our algorithms have lower False Positive Rate (FPR) and memory usage compared with the existing approaches to learned Bloom filter. We also demonstrated the improved performance of our algorithms on real-world datasets. |
| Tasks | |
| Published | 2020-01-01 |
| URL | https://openreview.net/forum?id=rJlNKCNtPB |
| https://openreview.net/pdf?id=rJlNKCNtPB | |
| PWC | https://paperswithcode.com/paper/adaptive-learned-bloom-filter-ada-bf-1 |
| Repo | |
| Framework | |
Disagreement-Regularized Imitation Learning
| Title | Disagreement-Regularized Imitation Learning |
| Authors | Anonymous |
| Abstract | We present a simple and effective algorithm designed to address the covariate shift problem in imitation learning. It operates by training an ensemble of policies on the expert demonstration data, and using the variance of their predictions as a cost which is minimized with RL together with a supervised behavioral cloning cost. Unlike adversarial imitation methods, it uses a fixed reward function which is easy to optimize. We prove a regret bound for the algorithm in the tabular setting which is linear in the time horizon multiplied by a coefficient which we show to be low for certain problems in which behavioral cloning fails. We evaluate our algorithm empirically across multiple pixel-based Atari environments and continuous control tasks, and show that it matches or significantly outperforms behavioral cloning and generative adversarial imitation learning. |
| Tasks | Continuous Control, Imitation Learning |
| Published | 2020-01-01 |
| URL | https://openreview.net/forum?id=rkgbYyHtwB |
| https://openreview.net/pdf?id=rkgbYyHtwB | |
| PWC | https://paperswithcode.com/paper/disagreement-regularized-imitation-learning |
| Repo | |
| Framework | |
Learning Generative Image Object Manipulations from Language Instructions
| Title | Learning Generative Image Object Manipulations from Language Instructions |
| Authors | Martin Längkvist, Andreas Persson, Amy Loutfi |
| Abstract | The use of adequate feature representations is essential for achieving high performance in high-level human cognitive tasks in computational modeling. Recent developments in deep convolutional and recurrent neural networks architectures enable learning powerful feature representations from both images and natural language text. Besides, other types of networks such as Relational Networks (RN) can learn relations between objects and Generative Adversarial Networks (GAN) have shown to generate realistic images. In this paper, we combine these four techniques to acquire a shared feature representation of the relation between objects in an input image and an object manipulation action description in the form of human language encodings to generate an image that shows the resulting end-effect the action would have on a computer-generated scene. The system is trained and evaluated on a simulated dataset and experimentally used on real-world photos. |
| Tasks | |
| Published | 2020-01-01 |
| URL | https://openreview.net/forum?id=rkgQL6VFwr |
| https://openreview.net/pdf?id=rkgQL6VFwr | |
| PWC | https://paperswithcode.com/paper/learning-generative-image-object |
| Repo | |
| Framework | |
Power up! Robust Graph Convolutional Network based on Graph Powering
| Title | Power up! Robust Graph Convolutional Network based on Graph Powering |
| Authors | Anonymous |
| Abstract | Graph convolutional networks (GCNs) are powerful tools for graph-structured data. However, they have been recently shown to be vulnerable to topological attacks. To enhance adversarial robustness, we go beyond spectral graph theory to robust graph theory. By challenging the classical graph Laplacian, we propose a new convolution operator that is provably robust in the spectral domain and is incorporated in the GCN architecture to improve expressivity and interpretability. By extending the original graph to a sequence of graphs, we also propose a robust training paradigm that encourages transferability across graphs that span a range of spatial and spectral characteristics. The proposed approaches are demonstrated in extensive experiments to {simultaneously} improve performance in both benign and adversarial situations. |
| Tasks | |
| Published | 2020-01-01 |
| URL | https://openreview.net/forum?id=BkxDxJHFDr |
| https://openreview.net/pdf?id=BkxDxJHFDr | |
| PWC | https://paperswithcode.com/paper/power-up-robust-graph-convolutional-network-1 |
| Repo | |
| Framework | |
Compressive Hyperspherical Energy Minimization
| Title | Compressive Hyperspherical Energy Minimization |
| Authors | Anonymous |
| Abstract | Minimum hyperspherical energy (MHE) has demonstrated its potential in regularizing neural networks and improving the generalization. MHE was inspired by the Thomson problem in physics where the distribution of multiple propelling electrons on a unit sphere can be modeled via minimizing some potential energy. Despite its practical effectiveness, MHE suffers from some difficulties in optimization as the dimensionality of the space becomes higher, therefore limiting the potential to improve network generalization. To address these problems, we propose the compressive minimum hyperspherical energy (CoMHE) as a more effective regularization for neural networks. Specifically, CoMHE utilizes a projection mapping to reduce the dimensionality of neurons and minimizes their hyperspherical energy. According to different constructions for the projection mapping, we propose two major variants: random projection CoMHE and angle-preserving CoMHE. As a novel extension, We further consider adversarial projection CoMHE and group CoMHE. We also provide some theoretical insights to justify the effectiveness. Our comprehensive experiments show that CoMHE consistently outperforms MHE by a considerable margin, and can be easily applied to improve different tasks such as image recognition and point cloud recognition. |
| Tasks | |
| Published | 2020-01-01 |
| URL | https://openreview.net/forum?id=BJgpDyHKwH |
| https://openreview.net/pdf?id=BJgpDyHKwH | |
| PWC | https://paperswithcode.com/paper/compressive-hyperspherical-energy-1 |
| Repo | |
| Framework | |
The Effect of Residual Architecture on the Per-Layer Gradient of Deep Networks
| Title | The Effect of Residual Architecture on the Per-Layer Gradient of Deep Networks |
| Authors | Anonymous |
| Abstract | A critical part of the training process of neural networks takes place in the very first gradient steps post initialization. In this work, we study the connection between the network’s architecture and initialization parameters, to the statistical properties of the gradient in random fully connected ReLU networks, through the study of the the Jacobian. We compare three types of architectures: vanilla networks, ResNets and DenseNets. The later two, as we show, preserve the variance of the gradient norm through arbitrary depths when initialized properly, which prevents exploding or decaying gradients at deeper layers. In addition, we show that the statistics of the per layer gradient norm is a function of the architecture and the layer’s size, but surprisingly not the layer’s depth. This depth invariant result is surprising in light of the literature results that state that the norm of the layer’s activations grows exponentially with the specific layer’s depth. Experimental support is given in order to validate our theoretical results and to reintroduce concatenated ReLU blocks, which, as we show, present better initialization properties than ReLU blocks in the case of fully connected networks. |
| Tasks | |
| Published | 2020-01-01 |
| URL | https://openreview.net/forum?id=H1eArT4tPH |
| https://openreview.net/pdf?id=H1eArT4tPH | |
| PWC | https://paperswithcode.com/paper/the-effect-of-residual-architecture-on-the |
| Repo | |
| Framework | |
Filling the Soap Bubbles: Efficient Black-Box Adversarial Certification with Non-Gaussian Smoothing
| Title | Filling the Soap Bubbles: Efficient Black-Box Adversarial Certification with Non-Gaussian Smoothing |
| Authors | Anonymous |
| Abstract | Randomized classifiers have been shown to provide a promising approach for achieving certified robustness against adversarial attacks in deep learning. However, most existing methods only leverage Gaussian smoothing noise and only work for $\ell_2$ perturbation. We propose a general framework of adversarial certification with non-Gaussian noise and for more general types of attacks, from a unified functional optimization perspective. Our new framework allows us to identify a key trade-off between accuracy and robustness via designing smoothing distributions, helping to design two new families of non-Gaussian smoothing distributions that work more efficiently for $\ell_2$ and $\ell_\infty$ attacks, respectively. Our proposed methods achieve better results than previous works and provide a new perspective on randomized smoothing certification. |
| Tasks | |
| Published | 2020-01-01 |
| URL | https://openreview.net/forum?id=Skg8gJBFvr |
| https://openreview.net/pdf?id=Skg8gJBFvr | |
| PWC | https://paperswithcode.com/paper/filling-the-soap-bubbles-efficient-black-box |
| Repo | |
| Framework | |
On learning visual odometry errors
| Title | On learning visual odometry errors |
| Authors | Anonymous |
| Abstract | This paper fosters the idea that deep learning methods can be sided to classical visual odometry pipelines to improve their accuracy and to produce uncertainty models to their estimations. We show that the biases inherent to the visual odom- etry process can be faithfully learnt and compensated for, and that a learning ar- chitecture associated to a probabilistic loss function can jointly estimate a full covariance matrix of the residual errors, defining a heteroscedastic error model. Experiments on autonomous driving image sequences and micro aerial vehicles camera acquisitions assess the possibility to concurrently improve visual odome- try and estimate an error associated to its outputs. |
| Tasks | Autonomous Driving, Visual Odometry |
| Published | 2020-01-01 |
| URL | https://openreview.net/forum?id=SklqvxSFDB |
| https://openreview.net/pdf?id=SklqvxSFDB | |
| PWC | https://paperswithcode.com/paper/on-learning-visual-odometry-errors |
| Repo | |
| Framework | |
Lagrangian Fluid Simulation with Continuous Convolutions
| Title | Lagrangian Fluid Simulation with Continuous Convolutions |
| Authors | Anonymous |
| Abstract | We present an approach to Lagrangian fluid simulation with a new type of convolutional network. Our networks process sets of moving particles, which describe fluids in space and time. Unlike previous approaches, we do not build an explicit graph structure to connect the particles but use spatial convolutions as the main differentiable operation that relates particles to their neighbors. To this end we present a simple, novel, and effective extension of N-D convolutions to the continuous domain. We show that our network architecture can simulate different materials, generalizes to arbitrary collision geometries, and can be used for inverse problems. In addition, we demonstrate that our continuous convolutions outperform prior formulations in terms of accuracy and speed. |
| Tasks | |
| Published | 2020-01-01 |
| URL | https://openreview.net/forum?id=B1lDoJSYDH |
| https://openreview.net/pdf?id=B1lDoJSYDH | |
| PWC | https://paperswithcode.com/paper/lagrangian-fluid-simulation-with-continuous |
| Repo | |
| Framework | |
Discriminative Particle Filter Reinforcement Learning for Complex Partial observations
| Title | Discriminative Particle Filter Reinforcement Learning for Complex Partial observations |
| Authors | Anonymous |
| Abstract | Deep reinforcement learning has succeeded in sophisticated games such as Atari, Go, etc. Real-world decision making, however, often requires reasoning with partial information extracted from complex visual observations. This paper presents Discriminative Particle Filter Reinforcement Learning (DPFRL), a new reinforcement learning framework for partial and complex observations. DPFRL encodes a differentiable particle filter with learned transition and observation models in a neural network, which allows for reasoning with partial observations over multiple time steps. While a standard particle filter relies on a generative observation model, DPFRL learns a discriminatively parameterized model that is training directly for decision making. We show that the discriminative parameterization results in significantly improved performance, especially for tasks with complex visual observations, because it circumvents the difficulty of modelling observations explicitly. In most cases, DPFRL outperforms state-of-the-art POMDP RL models in Flickering Atari Games, an existing POMDP RL benchmark, and in Natural Flickering Atari Games, a new, more challenging POMDP RL benchmark that we introduce. We further show that DPFRL performs well for visual navigation with real-world data. |
| Tasks | Atari Games, Decision Making, Visual Navigation |
| Published | 2020-01-01 |
| URL | https://openreview.net/forum?id=HJl8_eHYvS |
| https://openreview.net/pdf?id=HJl8_eHYvS | |
| PWC | https://paperswithcode.com/paper/discriminative-particle-filter-reinforcement |
| Repo | |
| Framework | |
Side-Tuning: Network Adaptation via Additive Side Networks
| Title | Side-Tuning: Network Adaptation via Additive Side Networks |
| Authors | Anonymous |
| Abstract | When training a neural network for a desired task, one may prefer to adapt a pretrained network rather than start with a randomly initialized one – due to lacking enough training data, performing lifelong learning where the system has to learn a new task while being previously trained for other tasks, or wishing to encode priors in the network via preset weights. The most commonly employed approaches for network adaptation are fine-tuning and using the pre-trained network as a fixed feature extractor, among others. In this paper we propose a straightforward alternative: Side-Tuning. Side-tuning adapts a pretrained network by training a lightweight “side” network that is fused with the (unchanged) pre-rained network using a simple additive process. This simple method works as well as or better than existing solutions while it resolves some of the basic issues with fine-tuning, fixed features, and several other common baselines. In particular, side-tuning is less prone to overfitting when little training data is available, yields better results than using a fixed feature extractor, and doesn’t suffer from catastrophic forgetting in lifelong learning. We demonstrate the performance of side-tuning under a diverse set of scenarios, including lifelong learning (iCIFAR, Taskonomy), reinforcement learning, imitation learning (visual navigation in Habitat), NLP question-answering (SQuAD v2), and single-task transfer learning (Taskonomy), with consistently promising results. |
| Tasks | Imitation Learning, Question Answering, Transfer Learning, Visual Navigation |
| Published | 2020-01-01 |
| URL | https://openreview.net/forum?id=HyxakgrFvS |
| https://openreview.net/pdf?id=HyxakgrFvS | |
| PWC | https://paperswithcode.com/paper/side-tuning-network-adaptation-via-additive |
| Repo | |
| Framework | |
NESTED LEARNING FOR MULTI-GRANULAR TASKS
| Title | NESTED LEARNING FOR MULTI-GRANULAR TASKS |
| Authors | Anonymous |
| Abstract | Standard deep neural networks (DNNs) used for classification are trained in an end-to-end fashion for very specific tasks - object recognition, face identification, character recognition, etc. This specificity often leads to overconfident models that generalize poorly to samples that are not from the original training distribution. Moreover, they do not allow to leverage information from heterogeneously annotated data, where for example, labels may be provided with different levels of granularity. Finally, standard DNNs do not produce results with simultaneous different levels of confidence for different levels of detail, they are most commonly an all or nothing approach. To address these challenges, we introduce the problem of nested learning: how to obtain a hierarchical representation of the input such that a coarse label can be extracted first, and sequentially refine this representation to obtain successively refined predictions, all of them with the corresponding confidence. We explicitly enforce this behaviour by creating a sequence of nested information bottlenecks. Looking at the problem of nested learning from an in formation theory perspective, we design a network topology with two important properties. First, a sequence of low dimensional (nested) feature embeddings are enforced. Then we show how the explicit combination of nested outputs can improve both robustness and finer predictions. Experimental results on CIFAR-10, MNIST, and FASHION-MNIST demonstrate that nested learning outperforms the same network trained in the standard end-to-end fashion. Since the network can be naturally trained with mixed data labeled at different levels of nested details, we also study what is the most efficient way of annotating data, when a fixed training budget is given and the cost of labels increases with the levels in the nested hierarchy. |
| Tasks | Face Identification, Object Recognition |
| Published | 2020-01-01 |
| URL | https://openreview.net/forum?id=Byxl-04KvH |
| https://openreview.net/pdf?id=Byxl-04KvH | |
| PWC | https://paperswithcode.com/paper/nested-learning-for-multi-granular-tasks |
| Repo | |
| Framework | |
What Can Neural Networks Reason About?
| Title | What Can Neural Networks Reason About? |
| Authors | Anonymous |
| Abstract | Neural networks have succeeded in many reasoning tasks. Empirically, these tasks require specialized network structures, e.g., Graph Neural Networks (GNNs) perform well on many such tasks, while less structured networks fail. Theoretically, there is limited understanding of why and when a network structure generalizes better than other equally expressive ones. We develop a framework to characterize which reasoning tasks a network can learn well, by studying how well its structure aligns with the algorithmic structure of the relevant reasoning procedure. We formally define algorithmic alignment and derive a sample complexity bound that decreases with better alignment. This framework explains the empirical success of popular reasoning models and suggests their limitations. We unify seemingly different reasoning tasks, such as intuitive physics, visual question answering, and shortest paths, via the lens of a powerful algorithmic paradigm, dynamic programming (DP). We show that GNNs can learn DP and thus solve these tasks. On several reasoning tasks, our theory aligns with empirical results. |
| Tasks | Question Answering, Visual Question Answering |
| Published | 2020-01-01 |
| URL | https://openreview.net/forum?id=rJxbJeHFPS |
| https://openreview.net/pdf?id=rJxbJeHFPS | |
| PWC | https://paperswithcode.com/paper/what-can-neural-networks-reason-about-1 |
| Repo | |
| Framework | |
Learning to Recognize the Unseen Visual Predicates
| Title | Learning to Recognize the Unseen Visual Predicates |
| Authors | Anonymous |
| Abstract | Visual relationship recognition models are limited in the ability to generalize from finite seen predicates to unseen ones. We propose a new problem setting named predicate zero-shot learning (PZSL): learning to recognize the predicates without training data. It is unlike the previous zero-shot learning problem on visual relationship recognition which learns to recognize the unseen relationship triplets (<subject, predicate, object>) but requires all components (subject, predicate, and object) to be seen in the training set. For the PZSL problem, however, the models are expected to recognize the diverse even unseen predicates, which is meaningful for many downstream high-level tasks, like visual question answering, to handle complex scenes and open questions. The PZSL is a very challenging task since the predicates are very abstract and follow an extreme long-tail distribution. To address the PZSL problem, we present a model that performs compatibility learning leveraging the linguistic priors from the corpus and knowledge base. An unbalanced sampled-softmax is further developed to tackle the extreme long-tail distribution of predicates. Finally, the experiments are conducted to analyze the problem and verify the effectiveness of our methods. The dataset and source code will be released for further study. |
| Tasks | Question Answering, Visual Question Answering, Zero-Shot Learning |
| Published | 2020-01-01 |
| URL | https://openreview.net/forum?id=rJecSyHtDS |
| https://openreview.net/pdf?id=rJecSyHtDS | |
| PWC | https://paperswithcode.com/paper/learning-to-recognize-the-unseen-visual |
| Repo | |
| Framework | |
