Deeper neural networks are more difficult to train. We present a residual learning framework to ease the training of networks that are substantially deeper than those used previously. We explicitly reformulate the layers as learning residual functions with reference to the layer inputs, instead of learning unreferenced functions. We provide comprehensive empirical evidence showing that these residual networks are easier to optimize, and can gain accuracy from considerably increased depth. On the ImageNet dataset we evaluate residual nets with a depth of up to 152 layers—8× deeper than VGG nets [40] but still having lower complexity. An ensemble of these residual nets achieves 3.57% error on the ImageNet test set. This result won the 1st place on the ILSVRC 2015 classification task. We also present analysis on CIFAR-10 with 100 and 1000 layers. The depth of representations is of central importance for many visual recognition tasks. Solely due to our extremely deep representations, we obtain a 28% relative improvement on the COCO object detection dataset. Deep residual nets are foundations of our submissions to ILSVRC & COCO 2015 competitions1, where we also won the 1st places on the tasks of ImageNet detection, ImageNet localization, COCO detection, and COCO segmentation.

/pdf/deep-residual-learning-for-image-recognition-b75jg61qrq.pdf

Deep Residual Learning for Image Recognition

We trained a large, deep convolutional neural network to classify the 1.2 million high-resolution images in the ImageNet LSVRC-2010 contest into the 1000 different classes. On the test data, we achieved top-1 and top-5 error rates of 37.5% and 17.0% which is considerably better than the previous state-of-the-art. The neural network, which has 60 million parameters and 650,000 neurons, consists of five convolutional layers, some of which are followed by max-pooling layers, and three fully-connected layers with a final 1000-way softmax. To make training faster, we used non-saturating neurons and a very efficient GPU implementation of the convolution operation. To reduce overriding in the fully-connected layers we employed a recently-developed regularization method called "dropout" that proved to be very effective. We also entered a variant of this model in the ILSVRC-2012 competition and achieved a winning top-5 test error rate of 15.3%, compared to 26.2% achieved by the second-best entry.

/pdf/imagenet-classification-with-deep-convolutional-neural-1190x9a6b9.pdf

ImageNet Classification with Deep Convolutional Neural Networks

In this work we investigate the effect of the convolutional network depth on its accuracy in the large-scale image recognition setting. Our main contribution is a thorough evaluation of networks of increasing depth using an architecture with very small (3x3) convolution filters, which shows that a significant improvement on the prior-art configurations can be achieved by pushing the depth to 16-19 weight layers. These findings were the basis of our ImageNet Challenge 2014 submission, where our team secured the first and the second places in the localisation and classification tracks respectively. We also show that our representations generalise well to other datasets, where they achieve state-of-the-art results. We have made our two best-performing ConvNet models publicly available to facilitate further research on the use of deep visual representations in computer vision.

Very Deep Convolutional Networks for Large-Scale Image Recognition

The explosion of image data on the Internet has the potential to foster more sophisticated and robust models and algorithms to index, retrieve, organize and interact with images and multimedia data. But exactly how such data can be harnessed and organized remains a critical problem. We introduce here a new database called “ImageNet”, a large-scale ontology of images built upon the backbone of the WordNet structure. ImageNet aims to populate the majority of the 80,000 synsets of WordNet with an average of 500-1000 clean and full resolution images. This will result in tens of millions of annotated images organized by the semantic hierarchy of WordNet. This paper offers a detailed analysis of ImageNet in its current state: 12 subtrees with 5247 synsets and 3.2 million images in total. We show that ImageNet is much larger in scale and diversity and much more accurate than the current image datasets. Constructing such a large-scale database is a challenging task. We describe the data collection scheme with Amazon Mechanical Turk. Lastly, we illustrate the usefulness of ImageNet through three simple applications in object recognition, image classification and automatic object clustering. We hope that the scale, accuracy, diversity and hierarchical structure of ImageNet can offer unparalleled opportunities to researchers in the computer vision community and beyond.

/pdf/imagenet-a-large-scale-hierarchical-image-database-1e8f663s1h.pdf

ImageNet: A large-scale hierarchical image database

We investigate the connection between two popular methods commonly used in 1 training deep neural networks: Sharpness-Aware Minimization (SAM) and Batch 2 Normalization. We find that perturbing only the affine BatchNorm parameters 3 in the adversarial step of SAM benefits the generalization performance, while 4 excluding them can decrease the performance strongly. We confirm our results 5 across several models and SAM-variants on CIFAR-10 and CIFAR-100 and show 6 preliminary results for ImageNet. Our results provide a practical tweak for training 7 deep networks, but also cast doubt on the commonly accepted explanation of SAM 8 minimizing a sharpness quantity responsible for generalization.

Perturbing BatchNorm and Only BatchNorm Benefits Sharpness-Aware Minimization

Deep reinforcement learning (RL) agents often fail to generalize beyond their training environments. To alleviate this problem, recent work has proposed the use of data augmentation. However, different tasks tend to benefit from different types of augmentations and selecting the right one typically requires expert knowledge. In this paper, we introduce three approaches for automatically finding an effective augmentation for any RL task. These are combined with two novel regularization terms for the policy and value function, required to make the use of data augmentation theoretically sound for actor-critic algorithms. We evaluate our method on the Procgen benchmark which consists of 16 procedurally generated environments and show that it improves test performance by 40% relative to standard RL algorithms. Our approach also outperforms methods specifically designed to improve generalization in RL, thus setting a new state-of-the-art on Procgen. In addition, our agent learns policies and representations which are more robust to changes in the environment that are irrelevant for solving the task, such as the background.

/pdf/automatic-data-augmentation-for-generalization-in-1nvqsnnx7k.pdf

Automatic Data Augmentation for Generalization in Reinforcement Learning

Standard deep reinforcement learning algorithms use a shared representation for the policy and value function. However, we argue that more information is needed to accurately estimate the value function than to learn the optimal policy. Consequently, the use of a shared representation for the policy and value function can lead to overfitting. To alleviate this problem, we propose two approaches which are combined to create IDAAC: Invariant Decoupled Advantage Actor-Critic. First, IDAAC decouples the optimization of the policy and value function, using separate networks to model them. Second, it introduces an auxiliary loss which encourages the representation to be invariant to task-irrelevant properties of the environment. IDAAC shows good generalization to unseen environments, achieving a new state-of-the-art on the Procgen benchmark and outperforming popular methods on DeepMind Control tasks with distractors. Moreover, IDAAC learns representations, value predictions, and policies that are more robust to aesthetic changes in the observations that do not change the underlying state of the environment.

Decoupling Value and Policy for Generalization in Reinforcement Learning

In the first part of this thesis, we develop novel image priors and efficient algorithms for image denoising and deconvolution applications. Our priors and algorithms enable fast, high-quality restoration of images corrupted by noise or blur. In the second part, we develop effective preconditioners for Laplacian matrices. Such matrices arise in a number of computer graphics and computational photography problems such as image colorization, tone mapping and geodesic distance computation on 3D meshes. 
The first prior we develop is a spectral prior which models correlations between different spectral bands. We introduce a prototype camera and flash system, used in conjunction with the spectral prior, to enable taking photographs at very low light levels. Our second prior is a sparsity-based measure for blind image deconvolution. This prior gives lower costs to sharp images than blurred ones, enabling the use simple and efficient Maximum a-Posteriori algorithms. 
We develop a new algorithm for the non-blind deconvolution problem. This enables extremely fast deconvolution of images blurred by a known blur kernel. Our algorithm uses Fast Fourier Transforms and Lookup Tables to achieve real-time deconvolution performance with non convex gradient-based priors. Finally, for certain image restoration problems with no known formation model, we demonstrate how learning a mapping between original/corrupted patch pairs enables effective restoration. 
The preconditioners we develop are multi-level schemes for discrete Poisson equations. Existing multilevel preconditioners have two major drawbacks: excessive bandwidth growth at coarse levels; and the inability to adapt to problems with highly varying coefficients. Our approach tackles both these problems by introducing sparsification and compensation steps at each level. We interleave the selection of fine and coarse-level variables with the removal of weak connections between potential fine-level variables (sparsification) and compensate for these changes by strengthening nearby connections. By applying these operations before each elimination step and repeating the procedure recursively on the resulting smaller systems, we obtain highly efficient schemes. The construction is linear in time and memory. Numerical experiments demonstrate that our new schemes outperform state of the art methods, both in terms of operation count and wall-clock time, over a range of 2D and 3D problems.

Low-level image priors and laplacian preconditioners for applications in computer graphics and computational photography

This paper formulates hypothesis verification as an RL problem Specifically, we aim to build an agent that, given a hypothesis about the dynamics of the world, can take actions to generate observations which can help predict whether the hypothesis is true or false Existing RL algorithms fail to solve this task, even for simple environments In order to train the agents, we exploit the underlying structure of many hypotheses, factorizing them as {pre-condition, action sequence, post-condition} triplets By leveraging this structure we show that RL agents are able to succeed at the task Furthermore, subsequent fine-tuning of the policies allows the agent to correctly verify hypotheses not amenable to the above factorization

/pdf/empirically-verifying-hypotheses-using-reinforcement-3oh0tjfltt.pdf

Rob Fergus

Papers

Perturbing BatchNorm and Only BatchNorm Benefits Sharpness-Aware Minimization

Automatic Data Augmentation for Generalization in Reinforcement Learning

Decoupling Value and Policy for Generalization in Reinforcement Learning

Low-level image priors and laplacian preconditioners for applications in computer graphics and computational photography

Empirically Verifying Hypotheses Using Reinforcement Learning