Paper Group NANR 128
Tree-structured Attention Module for Image Classification. Contrastive Multiview Coding. Training a Constrained Natural Media Painting Agent using Reinforcement Learning. Principled Weight Initialization for Hypernetworks. Frustratingly easy quasi-multitask learning. Improving Visual Relation Detection using Depth Maps. On Evaluating Explainability …
Tree-structured Attention Module for Image Classification
| Title | Tree-structured Attention Module for Image Classification |
| Authors | Anonymous |
| Abstract | Recent studies in attention modules have enabled higher performance in computer vision tasks by capturing global contexts and accordingly attending important features. In this paper, we propose a simple and highly parametrically efficient module named Tree-structured Attention Module (TAM) which recursively encourages neighboring channels to collaborate in order to produce a spatial attention map as an output. Unlike other attention modules which try to capture long-range dependencies at each channel, our module focuses on imposing non-linearities be- tween channels by utilizing point-wise group convolution. This module not only strengthens representational power of a model but also acts as a gate which controls signal flow. Our module allows a model to achieve higher performance in a highly parameter-efficient manner. We empirically validate the effectiveness of our module with extensive experiments on CIFAR-10/100 and SVHN datasets. With our proposed attention module employed, ResNet50 and ResNet101 models gain 2.3% and 1.2% accuracy improvement with less than 1.5% parameter over- head. Our PyTorch implementation code is publicly available. |
| Tasks | Image Classification |
| Published | 2020-01-01 |
| URL | https://openreview.net/forum?id=r1xBoxBYDH |
| https://openreview.net/pdf?id=r1xBoxBYDH | |
| PWC | https://paperswithcode.com/paper/tree-structured-attention-module-for-image |
| Repo | |
| Framework | |
Contrastive Multiview Coding
| Title | Contrastive Multiview Coding |
| Authors | Anonymous |
| Abstract | Humans view the world through many sensory channels, e.g., the long-wavelength light channel, viewed by the left eye, or the high-frequency vibrations channel, heard by the right ear. Each view is noisy and incomplete, but important factors, such as physics, geometry, and semantics, tend to be shared between all views (e.g., a “dog” can be seen, heard, and felt). We hypothesize that a powerful representation is one that models view-invariant factors. Based on this hypothesis, we investigate a contrastive coding scheme, in which a representation is learned that aims to maximize mutual information between different views but is otherwise compact. Our approach scales to any number of views, and is view-agnostic. The resulting learned representations perform above the state of the art for downstream tasks such as object classification, compared to formulations based on predictive learning or single view reconstruction, and improve as more views are added. On the Imagenet linear readoff benchmark, we achieve 68.4% top-1 accuracy. |
| Tasks | Object Classification |
| Published | 2020-01-01 |
| URL | https://openreview.net/forum?id=BkgStySKPB |
| https://openreview.net/pdf?id=BkgStySKPB | |
| PWC | https://paperswithcode.com/paper/contrastive-multiview-coding-1 |
| Repo | |
| Framework | |
Training a Constrained Natural Media Painting Agent using Reinforcement Learning
| Title | Training a Constrained Natural Media Painting Agent using Reinforcement Learning |
| Authors | Anonymous |
| Abstract | We present a novel approach to train a natural media painting using reinforcement learning. Given a reference image, our formulation is based on stroke-based rendering that imitates human drawing and can be learned from scratch without supervision. Our painting agent computes a sequence of actions that represent the primitive painting strokes. In order to ensure that the generated policy is predictable and controllable, we use a constrained learning method and train the painting agent using the environment model and follows the commands encoded in an observation. We have applied our approach on many benchmarks and our results demonstrate that our constrained agent can handle different painting media and different constraints in the action space to collaborate with humans or other agents. |
| Tasks | |
| Published | 2020-01-01 |
| URL | https://openreview.net/forum?id=SyeKGgStDB |
| https://openreview.net/pdf?id=SyeKGgStDB | |
| PWC | https://paperswithcode.com/paper/training-a-constrained-natural-media-painting |
| Repo | |
| Framework | |
Principled Weight Initialization for Hypernetworks
| Title | Principled Weight Initialization for Hypernetworks |
| Authors | Anonymous |
| Abstract | Hypernetworks are meta neural networks that generate weights for a main neural network in an end-to-end differentiable manner. Despite extensive applications ranging from multi-task learning to Bayesian deep learning, the problem of optimizing hypernetworks has not been studied to date. We observe that classical weight initialization methods like Glorot & Bengio (2010) and He et al. (2015), when applied directly on a hypernet, fail to produce weights for the mainnet in the correct scale. We develop principled techniques for weight initialization in hypernets, and show that they lead to more stable mainnet weights, lower training loss, and faster convergence. |
| Tasks | Multi-Task Learning |
| Published | 2020-01-01 |
| URL | https://openreview.net/forum?id=H1lma24tPB |
| https://openreview.net/pdf?id=H1lma24tPB | |
| PWC | https://paperswithcode.com/paper/principled-weight-initialization-for |
| Repo | |
| Framework | |
Frustratingly easy quasi-multitask learning
| Title | Frustratingly easy quasi-multitask learning |
| Authors | Anonymous |
| Abstract | We propose the technique of quasi-multitask learning (Q-MTL), a simple and easy to implement modification of standard multitask learning, in which the tasks to be modeled are identical. We illustrate it through a series of sequence labeling experiments over a diverse set of languages, that applying Q-MTL consistently increases the generalization ability of the applied models. The proposed architecture can be regarded as a new regularization technique encouraging the model to develop an internal representation of the problem at hand that is beneficial to multiple output units of the classifier at the same time. This property hampers the convergence to such internal representations which are highly specific and tailored for a classifier with a particular set of parameters. Our experiments corroborate that by relying on the proposed algorithm, we can approximate the quality of an ensemble of classifiers at a fraction of computational resources required. Additionally, our results suggest that Q-MTL handles the presence of noisy training labels better than ensembles. |
| Tasks | |
| Published | 2020-01-01 |
| URL | https://openreview.net/forum?id=H1lXCaVKvS |
| https://openreview.net/pdf?id=H1lXCaVKvS | |
| PWC | https://paperswithcode.com/paper/frustratingly-easy-quasi-multitask-learning |
| Repo | |
| Framework | |
Improving Visual Relation Detection using Depth Maps
| Title | Improving Visual Relation Detection using Depth Maps |
| Authors | Anonymous |
| Abstract | State of the art visual relation detection methods mostly rely on object information extracted from RGB images such as predicted class probabilities, 2D bounding boxes and feature maps. In this paper, we argue that the 3D positions of objects in space can provide additional valuable information about object relations. This information helps not only to detect spatial relations, such as \textit{standing behind}, but also non-spatial relations, such as \textit{holding}. Since 3D information of a scene is not easily accessible, we propose incorporating a pre-trained RGB-to-Depth model within visual relation detection frameworks. We discuss different feature extraction strategies from depth maps and show their critical role in relation detection. Our experiments confirm that the performance of state-of-the-art visual relation detection approaches can significantly be improved by utilizing depth map information. |
| Tasks | |
| Published | 2020-01-01 |
| URL | https://openreview.net/forum?id=HkxcUxrFPS |
| https://openreview.net/pdf?id=HkxcUxrFPS | |
| PWC | https://paperswithcode.com/paper/improving-visual-relation-detection-using-1 |
| Repo | |
| Framework | |
On Evaluating Explainability Algorithms
| Title | On Evaluating Explainability Algorithms |
| Authors | Anonymous |
| Abstract | A plethora of methods attempting to explain predictions of black-box models have been proposed by the Explainable Artificial Intelligence (XAI) community. Yet, measuring the quality of the generated explanations is largely unexplored, making quantitative comparisons non-trivial. In this work, we propose a suite of multifaceted metrics that enables us to objectively compare explainers based on the correctness, consistency, as well as the confidence of the generated explanations. These metrics are computationally inexpensive, do not require model-retraining and can be used across different data modalities. We evaluate them on common explainers such as Grad-CAM, SmoothGrad, LIME and Integrated Gradients. Our experiments show that the proposed metrics reflect qualitative observations reported in earlier works. |
| Tasks | |
| Published | 2020-01-01 |
| URL | https://openreview.net/forum?id=B1xBAA4FwH |
| https://openreview.net/pdf?id=B1xBAA4FwH | |
| PWC | https://paperswithcode.com/paper/on-evaluating-explainability-algorithms |
| Repo | |
| Framework | |
Using Hindsight to Anchor Past Knowledge in Continual Learning
| Title | Using Hindsight to Anchor Past Knowledge in Continual Learning |
| Authors | Anonymous |
| Abstract | In continual learning, the learner faces a stream of data whose distribution changes over time. Modern neural networks are known to suffer under this setting, as they quickly forget previously acquired knowledge. To address such catastrophic forgetting, state-of-the-art continual learning methods implement different types of experience replay, re-learning on past data stored in a small buffer known as episodic memory. In this work, we complement experience replay with a meta-learning technique that we call anchoring: the learner updates its knowledge on the current task, while keeping predictions on some anchor points of past tasks intact. These anchor points are learned using gradient-based optimization as to maximize forgetting of the current task, in hindsight, when the learner is fine-tuned on the episodic memory of past tasks. Experiments on several supervised learning benchmarks for continual learning demonstrate that our approach improves the state of the art in terms of both accuracy and forgetting metrics and for various sizes of episodic memories. |
| Tasks | Continual Learning, Meta-Learning |
| Published | 2020-01-01 |
| URL | https://openreview.net/forum?id=Hke12T4KPS |
| https://openreview.net/pdf?id=Hke12T4KPS | |
| PWC | https://paperswithcode.com/paper/using-hindsight-to-anchor-past-knowledge-in |
| Repo | |
| Framework | |
UNIVERSAL MODAL EMBEDDING OF DYNAMICS IN VIDEOS AND ITS APPLICATIONS
| Title | UNIVERSAL MODAL EMBEDDING OF DYNAMICS IN VIDEOS AND ITS APPLICATIONS |
| Authors | Anonymous |
| Abstract | Extracting underlying dynamics of objects in image sequences is one of the challenging problems in computer vision. On the other hand, dynamic mode decomposition (DMD) has recently attracted attention as a way of obtaining modal representations of nonlinear dynamics from (general multivariate time-series) data without explicit prior knowledge about the dynamics. In this paper, we propose a convolutional autoencoder based DMD (CAE-DMD) that is an extended DMD (EDMD) approach, to extract underlying dynamics in videos. To this end, we develop a modified CAE model by incorporating DMD on the encoder, which gives a more meaningful compressed representation of input image sequences. On the reconstruction side, a decoder is used to minimize the reconstruction error after applying the DMD, which in result gives an accurate reconstruction of inputs. We empirically investigated the performance of CAE-DMD in two applications: background/foreground extraction and video classification, on publicly available datasets. |
| Tasks | Time Series, Video Classification |
| Published | 2020-01-01 |
| URL | https://openreview.net/forum?id=H1lkYkrKDB |
| https://openreview.net/pdf?id=H1lkYkrKDB | |
| PWC | https://paperswithcode.com/paper/universal-modal-embedding-of-dynamics-in |
| Repo | |
| Framework | |
Continual Learning with Adaptive Weights (CLAW)
| Title | Continual Learning with Adaptive Weights (CLAW) |
| Authors | Anonymous |
| Abstract | Approaches to continual learning aim to successfully learn a set of related tasks that arrive in an online manner. Recently, several frameworks have been developed which enable deep learning to be deployed in this learning scenario. A key modelling decision is to what extent the architecture should be shared across tasks. On the one hand, separately modelling each task avoids catastrophic forgetting but it does not support transfer learning and leads to large models. On the other hand, rigidly specifying a shared component and a task-specific part enables task transfer and limits the model size, but it is vulnerable to catastrophic forgetting and restricts the form of task-transfer that can occur. Ideally, the network should adaptively identify which parts of the network to share in a data driven way. Here we introduce such an approach called Continual Learning with Adaptive Weights (CLAW), which is based on probabilistic modelling and variational inference. Experiments show that CLAW achieves state-of-the-art performance on six benchmarks in terms of overall continual learning performance, as measured by classification accuracy, and in terms of addressing catastrophic forgetting. |
| Tasks | Continual Learning, Transfer Learning |
| Published | 2020-01-01 |
| URL | https://openreview.net/forum?id=Hklso24Kwr |
| https://openreview.net/pdf?id=Hklso24Kwr | |
| PWC | https://paperswithcode.com/paper/continual-learning-with-adaptive-weights-claw |
| Repo | |
| Framework | |
Attacking Lifelong Learning Models with Gradient Reversion
| Title | Attacking Lifelong Learning Models with Gradient Reversion |
| Authors | Yunhui Guo, Mingrui Liu, Yandong Li, Liqiang Wang, Tianbao Yang, Tajana Rosing |
| Abstract | Lifelong learning aims at avoiding the catastrophic forgetting problem of traditional supervised learning models. Episodic memory based lifelong learning methods such as A-GEM (Chaudhry et al., 2018b) are shown to achieve the state-of-the-art results across the benchmarks. In A-GEM, a small episodic memory is utilized to store a random subset of the examples from previous tasks. While the model is trained on a new task, a reference gradient is computed on the episodic memory to guide the direction of the current update. While A-GEM has strong continual learning ability, it is not clear that if it can retain the performance in the presence of adversarial attacks. In this paper, we examine the robustness ofA-GEM against adversarial attacks to the examples in the episodic memory. We evaluate the effectiveness of traditional attack methods such as FGSM and PGD.The results show that A-GEM still possesses strong continual learning ability in the presence of adversarial examples in the memory and simple defense techniques such as label smoothing can further alleviate the adversarial effects. We presume that traditional attack methods are specially designed for standard supervised learning models rather than lifelong learning models. we therefore propose a principled way for attacking A-GEM called gradient reversion(GREV) which is shown to be more effective. Our results indicate that future lifelong learning research should bear adversarial attacks in mind to develop more robust lifelong learning algorithms. |
| Tasks | Continual Learning |
| Published | 2020-01-01 |
| URL | https://openreview.net/forum?id=SJlpy64tvB |
| https://openreview.net/pdf?id=SJlpy64tvB | |
| PWC | https://paperswithcode.com/paper/attacking-lifelong-learning-models-with |
| Repo | |
| Framework | |
Subjective Reinforcement Learning for Open Complex Environments
| Title | Subjective Reinforcement Learning for Open Complex Environments |
| Authors | Anonymous |
| Abstract | Solving tasks in open environments has been one of the long-time pursuits of reinforcement learning researches. We propose that data confusion is the core underlying problem. Although there exist methods that implicitly alleviate it from different perspectives, we argue that their solutions are based on task-specific prior knowledge that is constrained to certain kinds of tasks and lacks theoretical guarantees. In this paper, Subjective Reinforcement Learning Framework is proposed to state the problem from a broader and systematic view, and subjective policy is proposed to represent existing related algorithms in general. Theoretical analysis is given about the conditions for the superiority of a subjective policy, and the relationship between model complexity and the overall performance. Results are further applied as guidance for algorithm designing without task-specific prior knowledge about tasks. |
| Tasks | |
| Published | 2020-01-01 |
| URL | https://openreview.net/forum?id=S1eq9yrYvH |
| https://openreview.net/pdf?id=S1eq9yrYvH | |
| PWC | https://paperswithcode.com/paper/subjective-reinforcement-learning-for-open |
| Repo | |
| Framework | |
Learning To Explore Using Active Neural Mapping
| Title | Learning To Explore Using Active Neural Mapping |
| Authors | Anonymous |
| Abstract | This work presents a modular and hierarchical approach to learn policies for exploring 3D environments. Our approach leverages the strengths of both classical and learning-based methods, by using analytical path planners with learned mappers, and global and local policies. Use of learning provides flexibility with respect to input modalities (in mapper), leverages structural regularities of the world (in global policies), and provides robustness to errors in state estimation (in local policies). Such use of learning within each module retains its benefits, while at the same time, hierarchical decomposition and modular training allow us to sidestep the high sample complexities associated with training end-to-end policies. Our experiments in visually and physically realistic simulated 3D environments demonstrate the effectiveness of our proposed approach over past learning and geometry-based approaches. |
| Tasks | |
| Published | 2020-01-01 |
| URL | https://openreview.net/forum?id=HklXn1BKDH |
| https://openreview.net/pdf?id=HklXn1BKDH | |
| PWC | https://paperswithcode.com/paper/learning-to-explore-using-active-neural |
| Repo | |
| Framework | |
Crafting Data-free Universal Adversaries with Dilate Loss
| Title | Crafting Data-free Universal Adversaries with Dilate Loss |
| Authors | Anonymous |
| Abstract | We introduce a method to create Universal Adversarial Perturbations (UAP) for a given CNN in a data-free manner. Data-free approaches suite scenarios where the original training data is unavailable for crafting adversaries. We show that the adversary generation with full training data can be approximated to a formulation without data. This is realized through a sequential optimization of the adversarial perturbation with the proposed dilate loss. Dilate loss basically maximizes the Euclidean norm of the output before nonlinearity at any layer. By doing so, the perturbation constrains the ReLU activation function at every layer to act roughly linear for data points and thus eliminate the dependency on data for crafting UAPs. Extensive experiments demonstrate that our method not only has theoretical support, but achieves higher fooling rate than the existing data-free work. Furthermore, we evidence improvement in limited data cases as well. |
| Tasks | |
| Published | 2020-01-01 |
| URL | https://openreview.net/forum?id=HJxVC1SYwr |
| https://openreview.net/pdf?id=HJxVC1SYwr | |
| PWC | https://paperswithcode.com/paper/crafting-data-free-universal-adversaries-with |
| Repo | |
| Framework | |
On Iterative Neural Network Pruning, Reinitialization, and the Similarity of Masks
| Title | On Iterative Neural Network Pruning, Reinitialization, and the Similarity of Masks |
| Authors | Anonymous |
| Abstract | We examine how recently documented, fundamental phenomena in deep learn-ing models subject to pruning are affected by changes in the pruning procedure. Specifically, we analyze differences in the connectivity structure and learning dynamics of pruned models found through a set of common iterative pruning techniques, to address questions of uniqueness of trainable, high-sparsity sub-networks, and their dependence on the chosen pruning method. In convolutional layers, we document the emergence of structure induced by magnitude-based un-structured pruning in conjunction with weight rewinding that resembles the effects of structured pruning. We also show empirical evidence that weight stability can be automatically achieved through apposite pruning techniques. |
| Tasks | Network Pruning |
| Published | 2020-01-01 |
| URL | https://openreview.net/forum?id=B1xgQkrYwS |
| https://openreview.net/pdf?id=B1xgQkrYwS | |
| PWC | https://paperswithcode.com/paper/on-iterative-neural-network-pruning |
| Repo | |
| Framework | |
