While deep reinforcement learning techniques have led to agents that are successfully able to learn to perform a number of tasks that had been previously unlearnable, these techniques are still susceptible to the longstanding problem of reward sparsity. This is especially true for tasks such as training an agent to play StarCraft II, a real-time strategy game where reward is only given at the end of a game which is usually very long. While this problem can be addressed through reward shaping, such approaches typically require a human expert with specialized knowledge. Inspired by the vision of enabling reward shaping through the more-accessible paradigm of natural-language narration, we develop a technique that can provide the benefits of reward shaping using natural language commands. Our narration-guided RL agent projects sequences of natural-language commands into the same high-dimensional representation space as corresponding goal states. We show that we can get improved performance with our method compared to traditional reward-shaping approaches. Additionally, we demonstrate the ability of our method to generalize to unseen natural-language commands.

Using generative models for Inverse Graphics is an active area of research. However, most works focus on developing models for supervised and semi-supervised methods. In this paper, we study the problem of unsupervised learning of 3D geometry from single images. Our approach is to use a generative model that produces 2-D images as projections of a latent 3D voxel grid, which we train either as a variational auto-encoder or using adversarial methods. Our contributions are as follows: First, we show how to recover 3D shape and pose from general datasets such as MNIST, and MNIST Fashion in good quality. Second, we compare the shapes learned using adversarial and variational methods. Adversarial approach gives denser 3D shapes. Third, we explore the idea of modelling the pose of an object as uniform distribution to recover 3D shape from a single image. Our experiment with the CelebA dataset \cite{liu2015faceattributes} proves that we can recover complete 3D shape from a single image when the object is symmetric along one, or more axis whilst results obtained using ModelNet40 \cite{wu20153d} show the potential side-effects, in which the model learns 3D shapes such that it can render the same image from any viewpoint. Forth, we present a general end-to-end approach to learning 3D shapes from single images in a completely unsupervised fashion by modelling the factors of variation such as azimuth as independent latent variables. Our method makes no assumptions about the dataset, and can work with synthetic as well as real images (i.e. unsupervised in true sense). We present our results, by training the model using the $\mu$-VAE objective \cite{ucar2019bridging} and a dataset combining all images from MNIST, MNIST Fashion, CelebA and six categories of ModelNet40. The model is able to learn 3D shapes and the pose in qood quality and leverages information learned across all datasets.

This work proposes a novel neural network architecture, called the Dynamically Controlled Recurrent Neural Network (DCRNN), specifically designed to model dynamical systems that are governed by ordinary differential equations (ODEs). The current state vectors of these types of dynamical systems only depend on their state-space models, along with the respective inputs and initial conditions. Long Short-Term Memory (LSTM) networks, which have proven to be very effective for memory-based tasks, may fail to model physical processes as they tend to memorize, rather than learn how to capture the information on the underlying dynamics. The proposed DCRNN includes learnable skip-connections across previously hidden states, and introduces a regularization term in the loss function by relying on Lyapunov stability theory. The regularizer enables the placement of eigenvalues of the transfer function induced by the DCRNN to desired values, thereby acting as an internal controller for the hidden state trajectory. The results show that, for forecasting a chaotic dynamical system, the DCRNN outperforms the LSTM in $100$ out of $100$ randomized experiments by reducing the mean squared error of the LSTM's forecasting by $80.0\% \pm 3.0\%$.

In this paper, we study the problem of learning probabilistic logical rules for inductive and interpretable link prediction. Despite the importance of inductive link prediction, most previous works focused on transductive link prediction and cannot manage previously unseen entities. Moreover, they are black-box models that are not easily explainable for humans. We propose DRUM, a scalable and differentiable approach for mining first-order logical rules from knowledge graphs which resolves these problems. We motivate our method by making a connection between learning confidence scores for each rule and low-rank tensor approximation. DRUM uses bidirectional RNNs to share useful information across the tasks of learning rules for different relations. We also empirically demonstrate the efficiency of DRUM over existing rule mining methods for inductive link prediction on a variety of benchmark datasets.

Single-agent deep reinforcement learning has achieved impressive performance in many domains, including playing Go [1, 2] and Atari games [3, 4]. However, many real world problems, such as traffic congestion reduction [5, 6], antenna tilt control [7], and dynamic resource allocation [8] are more naturally modeled as multi-agent systems. Unfortunately, directly deploying single-agent reinforcement learning to each agent in a multi-agent system does not result in satisfying performance [9, 10]. Particularly, in multi-agent reinforcement learning [8, 10-19], estimating the value function is challenging, because the environment is non-stationary from the perspective of an individual agent [10, 11]. To alleviate the issue, recently, multi-agent deep deterministic policy gradient (MADDPG) [10] proposed a centralized critic whose input is the concatenation of all agents' observations and actions.

Inferring the most likely configuration for a subset of variables of a joint distribution given the remaining ones - which we refer to as co-generation - is an important challenge that is computationally demanding for all but the simplest settings. This task has received a considerable amount of attention, particularly for classical ways of modeling distributions like structured prediction. In contrast, almost nothing is known about this task when considering recently proposed techniques for modeling high-dimensional distributions, particularly generative adversarial nets (GANs). Therefore, in this paper, we study the occurring challenges for co-generation with GANs. To address those challenges we develop an annealed importance sampling based Hamiltonian Monte Carlo co-generation algorithm. The presented approach significantly outperforms classical gradient based methods on a synthetic and on the CelebA and LSUN datasets.

For joint inference over multiple variables, a variety of structured prediction techniques have been developed to model correlations among variables and thereby improve predictions. However, many classical approaches suffer from one of two primary drawbacks: they either lack the ability to model high-order correlations among variables while maintaining computationally tractable inference, or they do not allow to explicitly model known correlations. To address this shortcoming, we introduce `Graph Structured Prediction Energy Networks,' for which we develop inference techniques that allow to both model explicit local and implicit higher-order correlations while maintaining tractability of inference. We apply the proposed method to tasks from the natural language processing and computer vision domain and demonstrate its general utility.

We rerank with scores from pretrained masked language models like BERT to improve ASR and NMT performance. These log-pseudolikelihood scores (LPLs) can outperform large, autoregressive language models (GPT -2) in out-of-the-box scoring. RoBERTa reduces WER by up to 30% relative on an end-to-end LibriSpeech system and adds up to 1.7 BLEU on state-of-the-art baselines for TED Talks low-resource pairs, with further gains from domain adaptation. In the multilingual setting, a single XLM can be used to rerank translation outputs in multiple languages. The numerical and qualitative properties of LPL scores suggest that LPLs capture sentence fluency better than autoregressive scores. Finally, we finetune BERT to estimate sentence LPLs without masking, enabling scoring in a single, non-recurrent inference pass.

We propose an algorithm to enhance certified robustness of a deep model ensemble by optimally weighting each base model. Unlike previous works on using ensembles to empirically improve robustness, our algorithm is based on optimizing a guaranteed robustness certificate of neural networks. Our proposed ensemble framework with certified robustness, RobBoost, formulates the optimal model selection and weighting task as an optimization problem on a lower bound of classification margin, which can be efficiently solved using coordinate descent. Experiments show that our algorithm can form a more robust ensemble than naively averaging all available models using robustly trained MNIST or CIFAR base models. Additionally, our ensemble typically has better accuracy on clean (unperturbed) data. RobBoost allows us to further improve certified robustness and clean accuracy by creating an ensemble of already certified models.

Denoising and Regularization via Exploiting the Structural Bias of Convolutional Generators Reinhard Heckel and Mahdi Soltanolkotabi † Dept. of Electrical and Computer Engineering, Technical University of Munich † Dept. of Electrical and Computer Engineering, University of Southern California November 1, 2019 Abstract Convolutional Neural Networks (CNNs) have emerged as highly successful tools for image generation, recovery, and restoration. This success is often attributed to large amounts of training data. However, recent experimental findings challenge this view and instead suggest that a major contributing factor to this success is that convolutional networks impose strong prior assumptions about natural images. A surprising experiment that highlights this architectural bias towards natural images is that one can remove noise and corruptions from a natural image without using any training data, by simply fitting (via gradient descent) a randomly initialized, over-parameterized convolutional generator to the single corrupted image. While this over-parameterized network can fit the corrupted image perfectly, surprisingly after a few iterations of gradient descent one obtains the uncorrupted image. This intriguing phenomena enables state-of-the-art CNN-based denoising and regularization of linear inverse problems such as compressive sensing. In this paper we take a step towards demystifying this experimental phenomena by attributing this effect to particular architectural choices of convolutional networks, namely convolutions with fixed interpolating filters. We then formally characterize the dynamics of fitting a two layer convolutional generator to a noisy signal and prove that early-stopped gradient descent denoises/regularizes. This results relies on showing that convolutional generators fit the structured part of an image significantly faster than the corrupted portion. 1 Introduction Convolutional neural networks are extremely popular for image generation. The majority of image generating networks is convolutional, ranging from Deep Convolutional Generative Adversarial Networks (DC-GANs) [Rad 15] to the U-Net [Ron 15]. It is well known that convolutional neural networks incorporate implicit assumption about the signals they generate, such as pixels that are close being related. This makes them particularly well suited for representing sets of images or modeling distributions of images. It is less known, however, that those prior assumptions build into the architecture are so strong that convolutional neural networks are useful even without ever being exposed to training data. The latter was first shown in the Deep Image Prior (DIP) paper [Uly 18]. Ulyanov et al. [Uly 18] observed that when'training' an standard convolutional auto-encoder such as the popular U-net [Ron 15] on a single noisy image and regularizing by early stopping, the network performs image restoration such as denoising with state-of-the-art performance.