We present Optimal Completion Distillation (OCD), a training procedure for optimizing sequence to sequence models based on edit distance. OCD is efficient, has no hyper-parameters of its own, and does not require pretraining or joint optimization with conditional log-likelihood. Given a partial sequence generated by the model, we first identify the set of optimal suffixes that minimize the total edit distance, using an efficient dynamic programming algorithm. Then, for each position of the generated sequence, we use a target distribution that puts equal probability on the first token of all the optimal suffixes. OCD achieves the state-of-the-art performance on end-to-end speech recognition, on both Wall Street Journal and Librispeech datasets, achieving $9.3\%$ WER and $4.5\%$ WER respectively.

Humans learn complex latent structures from their environments (e.g., natural language, mathematics, music, social hierarchies). In cognitive science and cognitive neuroscience, models that infer higher-order structures from sensory or first-order representations have been proposed to account for the complexity and flexibility of human behavior. But how do the structures that these models invoke arise in neural systems in the first place? To answer this question, we explain how a system can learn latent representational structures (i.e., predicates) from experience with wholly unstructured data. During the process of predicate learning, an artificial neural network exploits the naturally occurring dynamic properties of distributed computing across neuronal assemblies in order to learn predicates, but also to combine them compositionally, two computational aspects which appear to be necessary for human behavior as per formal theories in multiple domains. We describe how predicates can be combined generatively using neural oscillations to achieve human-like extrapolation and compositionality in an artificial neural network. The ability to learn predicates from experience, to represent structures compositionally, and to extrapolate to unseen data offers an inroads to understanding and modeling the most complex human behaviors.

Deep neural networks, in particular convolutional neural networks, have become highly effective tools for compressing images and solving inverse problems including denoising, inpainting, and reconstruction from few and noisy measurements. This success can be attributed in part to their ability to represent and generate natural images well. Contrary to classical tools such as wavelets, image-generating deep neural networks have a large number of parameters---typically a multiple of their output dimension---and need to be trained on large datasets. In this paper, we propose an untrained simple image model, called the deep decoder, which is a deep neural network that can generate natural images from very few weight parameters. The deep decoder has a simple architecture with no convolutions and fewer weight parameters than the output dimensionality. This underparameterization enables the deep decoder to compress images into a concise set of network weights, which we show is on par with wavelet-based thresholding. Further, underparameterization provides a barrier to overfitting, allowing the deep decoder to have state-of-the-art performance for denoising. The deep decoder is simple in the sense that each layer has an identical structure that consists of only one upsampling unit, pixel-wise linear combination of channels, ReLU activation, and channelwise normalization. This simplicity makes the network amenable to theoretical analysis, and it sheds light on the aspects of neural networks that enable them to form effective signal representations.

In this brief technical report we introduce the CINIC-10 dataset as a plug-in extended alternative for CIFAR-10. It was compiled by combining CIFAR-10 with images selected and downsampled from the ImageNet database. We present the approach to compiling the dataset, illustrate the example images for different classes, give pixel distributions for each part of the repository, and give some standard benchmarks for well known models. Details for download, usage, and compilation can be found in the associated github repository.

Estimating fundamental matrices is a classic problem in computer vision. Traditional methods rely heavily on the correctness of estimated key-point correspondences, which can be noisy and unreliable. As a result, it is difficult for these methods to handle image pairs with large occlusion or significantly different camera poses. In this paper, we propose novel neural network architectures to estimate fundamental matrices in an end-to-end manner without relying on point correspondences. New modules and layers are introduced in order to preserve mathematical properties of the fundamental matrix as a homogeneous rank-2 matrix with seven degrees of freedom. We analyze performance of the proposed models using various metrics on the KITTI dataset, and show that they achieve competitive performance with traditional methods without the need for extracting correspondences.

Real-life control tasks involve matter of various substances--rigid or soft bodies, liquid, gas--each with distinct physical behaviors. This poses challenges to traditional rigid-body physics engines. Particle-based simulators have been developed to model the dynamics of these complex scenes; however, relying on approximation techniques, their simulation often deviates from real world physics, especially in the long term. In this paper, we propose to learn a particle-based simulator for complex control tasks. Combining learning with particle-based systems brings in two major benefits: first, the learned simulator, just like other particle-based systems, acts widely on objects of different materials; second, the particle-based representation poses strong inductive bias for learning: particles of the same type have the same dynamics within. This enables the model to quickly adapt to new environments of unknown dynamics within a few observations. Using the learned simulator, robots have achieved success in complex manipulation tasks, such as manipulating fluids and deformable foam. The effectiveness of our method has also been demonstrated in the real world. Our study helps lay the foundation for robot learning of dynamic scenes with particle-based representations. Under the same push, a rigid box will slide, modeling clay will deform, and a cup full of water will fall with water spilling out.

Computational identification of promoters is notoriously difficult as human genes often have unique promoter sequences that provide regulation of transcription and interaction with transcription initiation complex. While there are many attempts to develop computational promoter identification methods, we have no reliable tool to analyze long genomic sequences. In this work we further develop our deep learning approach that was relatively successful to discriminate short promoter and non-promoter sequences. Instead of focusing on the classification accuracy, in this work we predict the exact positions of the TSS inside the genomic sequences testing every possible location. We studied human promoters to find effective regions for discrimination and built corresponding deep learning models. These models use adaptively constructed negative set which iteratively improves the models discriminative ability. The developed promoter identification models significantly outperform the previously developed promoter prediction programs by considerably reducing the number of false positive predictions. The best model we have built has recall 0.76, precision 0.77 and MCC 0.76, while the next best tool FPROM achieved precision 0.48 and MCC 0.60 for the recall of 0.75. Our method is available at http://www.cbrc.kaust.edu.sa/PromID/.

Modern data analysis pipelines are becoming increasingly complex due to the presence of multi-view information sources. While graphs are effective in modeling complex relationships, in many scenarios a single graph is rarely sufficient to succinctly represent all interactions, and hence multi-layered graphs have become popular. Though this leads to richer representations, extending solutions from the single-graph case is not straightforward. Consequently, there is a strong need for novel solutions to solve classical problems, such as node classification, in the multi-layered case. In this paper, we consider the problem of semi-supervised learning with multi-layered graphs. Though deep network embeddings, e.g. DeepWalk, are widely adopted for community discovery, we argue that feature learning with random node attributes, using graph neural networks, can be more effective. To this end, we propose to use attention models for effective feature learning, and develop two novel architectures, GrAMME-SG and GrAMME-Fusion, that exploit the inter-layer dependencies for building multi-layered graph embeddings. Using empirical studies on several benchmark datasets, we evaluate the proposed approaches and demonstrate significant performance improvements in comparison to state-of-the-art network embedding strategies. The results also show that using simple random features is an effective choice, even in cases where explicit node attributes are not available.

Deep learning based speech enhancement and source separation systems have recently reached unprecedented levels of quality, to the point that performance is reaching a new ceiling. Most systems rely on estimating the magnitude of a target source by estimating a real-valued mask to be applied to a time-frequency representation of the mixture signal. A limiting factor in such approaches is a lack of phase estimation: the phase of the mixture is most often used when reconstructing the estimated time-domain signal. Here, we propose `MagBook', `phasebook', and `Combook', three new types of layers based on discrete representations that can be used to estimate complex time-frequency masks. MagBook layers extend classical sigmoidal units and a recently introduced convex softmax activation for mask-based magnitude estimation. Phasebook layers use a similar structure to give an estimate of the phase mask without suffering from phase wrapping issues. Combook layers are an alternative to the MagBook-Phasebook combination that directly estimate complex masks. We present various training and inference regimes involving these representations, and explain in particular how to include them in an end-to-end learning framework. We also present an oracle study to assess upper bounds on performance for various types of masks using discrete phase representations. We evaluate the proposed methods on the wsj0-2mix dataset, a well-studied corpus for single-channel speaker-independent speaker separation, matching the performance of state-of-the-art mask-based approaches without requiring additional phase reconstruction steps.

Recently, Convolution Neural Networks (CNNs) obtained huge success in numerous vision tasks. In particular, DenseNets have demonstrated that feature reuse via dense skip connections can effectively alleviate the difficulty of training very deep networks and that reusing features generated by the initial layers in all subsequent layers has strong impact on performance. To feed even richer information into the network, a novel adaptive Multi-scale Convolution Aggregation module is presented in this paper. Composed of layers for multi-scale convolutions, trainable cross-scale aggregation, maxout, and concatenation, this module is highly non-linear and can boost the accuracy of DenseNet while using much fewer parameters. In addition, due to high model complexity, the network with extremely dense feature reuse is prone to overfitting. To address this problem, a regularization method named Stochastic Feature Reuse is also presented. Through randomly dropping a set of feature maps to be reused for each mini-batch during the training phase, this regularization method reduces training costs and prevents co-adaptation. Experimental results on CIFAR-10, CIFAR-100 and SVHN benchmarks demonstrated the effectiveness of the proposed methods.