Explore the latest research from Facebook

All Publications

December 13, 2019 David Novotny, Benjamin Graham, Jeremy Reizenstein
Paper

PerspectiveNet: A Scene-consistent Image Generator for New View Synthesis in Real Indoor Environments

Given a set of a reference RGBD views of an indoor environment, and a new viewpoint, our goal is to predict the view from that location. Prior work on new-view generation has predominantly focused on significantly constrained scenarios, typically involving artificially rendered views of isolated CAD models. Here we tackle a much more challenging version of the problem. We devise an approach that exploits known geometric properties of the scene (per-frame camera extrinsics and depth) in order to warp reference views into the new ones.
Paper
December 12, 2019 Brenden Lake
Paper

Compositional generalization through meta sequence-to-sequence learning

People can learn a new concept and use it compositionally, understanding how to “blicket twice” after learning how to “blicket.” In contrast, powerful sequence-to-sequence (seq2seq) neural networks fail such tests of compositionality, especially when composing new concepts together with existing concepts. In this paper, I show how memory-augmented neural networks can be trained to generalize compositionally through meta seq2seq learning.
Paper
December 11, 2019 Eliya Nachmani, Lior Wolf
Paper

Hyper-Graph-Network Decoders for Block Codes

Neural decoders were shown to outperform classical message passing techniques for short BCH codes. In this work, we extend these results to much larger families of algebraic block codes, by performing message passing with graph neural networks.
Paper
December 11, 2019 Stéphane d'Ascoli, Levent Sagun, Joan Bruna, Giulio Biroli
Paper

Finding the Needle in the Haystack with Convolutions: on the benefits of architectural bias

Despite the phenomenal success of deep neural networks in a broad range of learning tasks, there is a lack of theory to understand the way they work. In particular, Convolutional Neural Networks (CNNs) are known to perform much better than Fully-Connected Networks (FCNs) on spatially structured data: the architectural structure of CNNs benefits from prior knowledge on the features of the data, for instance their translation invariance. The aim of this work is to understand this fact through the lens of dynamics in the loss landscape.
Paper
December 10, 2019 Anton Bakhtin, Laurens van der Maaten, Justin Johnson, Laura Gustafson, Ross Girshick
Paper

PHYRE: A New Benchmark for Physical Reasoning

Understanding and reasoning about physics is an important ability of intelligent agents. We develop the PHYRE benchmark for physical reasoning that contains a set of simple classical mechanics puzzles in a 2D physical environment.
Paper
December 10, 2019 Andrea Zanette, Alessandro Lazaric, Mykel J. Kochenderfer, Emma Brunskill
Paper

Limiting Extrapolation in Linear Approximate Value Iteration

We study linear approximate value iteration (LAVI) with a generative model. While linear models may accurately represent the optimal value function using a few parameters, several empirical and theoretical studies show the combination of least-squares projection with the Bellman operator may be expansive, thus leading LAVI to amplify errors over iterations and eventually diverge. We introduce an algorithm that approximates value functions by combining Q-values estimated at a set of anchor states.
Paper
December 10, 2019 Ari Morcos, Haonan Yu, Michela Paganini, Yuandong Tian
Paper

One ticket to win them all: generalizing lottery ticket initializations across datasets and optimizers

The success of lottery ticket initializations [7] suggests that small, sparsified networks can be trained so long as the network is initialized appropriately. Unfortunately, finding these “winning ticket” initializations is computationally expensive. One potential solution is to reuse the same winning tickets across a variety of datasets and optimizers. However, the generality of winning ticket initializations remains unclear. Here, we attempt to answer this question by generating winning tickets for one training configuration (optimizer and dataset) and evaluating their performance on another configuration.
Paper
December 10, 2019 Alexander Peysakhovich, Christian Kroer, Adam Lerer
Paper

Robust Multi-agent Counterfactual Prediction

We consider the problem of using logged data to make predictions about what would happen if we changed the ‘rules of the game’ in a multi-agent system. This task is difficult because in many cases we observe actions individuals take but not their private information or their full reward functions. In addition, agents are strategic, so when the rules change, they will also change their actions.
Paper
December 9, 2019 Hugo Touvron, Andrea Vedaldi, Matthijs Douze, Hervé Jégou
Paper

Fixing the train-test resolution discrepancy

Data-augmentation is key to the training of neural networks for image classification. This paper first shows that existing augmentations induce a significant discrepancy between the size of the objects seen by the classifier at train and test time: in fact, a lower train resolution improves the classification at test time!
Paper
December 9, 2019 Mahmoud Assran, Joshua Romoff, Nicolas Ballas, Joelle Pineau, Michael G. Rabbat
Paper

Gossip-based Actor-Learner Architectures for Deep Reinforcement Learning

Multi-simulator training has contributed to the recent success of Deep Reinforcement Learning by stabilizing learning and allowing for higher training throughputs. We propose Gossip-based Actor-Learner Architectures (GALA) where several actor-learners (such as A2C agents) are organized in a peer-to-peer communication topology, and exchange information through asynchronous gossip in order to take advantage of a large number of distributed simulators.
Paper