Publications

Publication

February 14, 2020

Uncertainty-guided Continual Learning with Bayesian Neural Networks | Facebook AI Research

Continual learning aims to learn new tasks without forgetting previously learned ones. This is especially challenging when one cannot access data from previous tasks and when the model has a fixed capacity. Current regularization-based…

Sayna Ebrahimi, Mohamed Elhoseiny, Trevor Darrell, Marcus Rohrbach

February 14, 2020

Download the Paper

Publication

February 07, 2020

INTEGRITY

Generate, Segment and Refine: Towards Generic Manipulation Segmentation | Facebook AI Research

Detecting manipulated images has become a significant emerging challenge. The advent of image sharing platforms and the easy availability of advanced photo editing software have resulted in a large quantities of manipulated images being shared…

Peng Zhou, Bor-Chun Chen, Xintong Han, Mahyar Najibi, Abhinav Shrivastava, Ser-Nam Lim, Larry S. Davis

February 07, 2020

Read the Paper

Publication

February 07, 2020

FASTER Recurrent Networks for Efficient Video Classification | Facebook AI Research

Typical video classification methods often divide a video into short clips, do inference on each clip independently, then aggregate the clip-level predictions to generate the video-level results. However, processing visually similar clips…

Linchao Zhu, Du Tran, Laura Sevilla-Lara, Yi Yang, Matt Feiszli, Heng Wang

February 07, 2020

Read the Paper

Publication

January 13, 2020

NLP

Scaling up online speech recognition using ConvNets | Facebook AI Research

We design an online end-to-end speech recognition system based on Time-Depth Separable (TDS) convolutions and Connectionist Temporal Classification (CTC). The system has almost three times the throughput of a well tuned hybrid ASR baseline…

Vineel Pratap, Qiantong Xu, Jacob Kahn, Gilad Avidov, Tatiana Likhomanenko, Awni Hannun, Vitaliy Liptchinsky, Gabriel Synnaeve, Ronan Collobert

January 13, 2020

Read the Paper

Publication

December 14, 2019

From Senones to Chenones: Tied Context-Dependent Graphemes for Hybrid Speech Recognition | Facebook AI Research

There is an implicit assumption that traditional hybrid approaches for automatic speech recognition (ASR) cannot directly model graphemes and need to rely on phonetic lexicons to get competitive performance, especially on English which has poor…

Duc Le, Xiaohui Zhang, Weiyi Zhang, Christian Fuegen, Geoffrey Zweig, Michael L. Seltzer

December 14, 2019

Read the Paper

Publication

December 14, 2019

NLP

From Senones to Chenones: Tied Context-Dependent Graphemes for Hybrid Speech Recognition | Facebook AI Research

Duc Le, Xiaohui Zhang, Weiyi Zhang, Christian Fuegen, Geoffrey Zweig, Michael L. Seltzer

December 14, 2019

Read the Paper

Publication

December 13, 2019

COMPUTER VISION

SPEECH & AUDIO

Lead2Gold: Towards exploiting the full potential of noisy transcriptions for speech recognition | Facebook AI Research

The transcriptions used to train an Automatic Speech Recognition (ASR) system may contain errors. Usually, either a quality control stage discards transcriptions with too many errors, or the noisy transcriptions are used as is. We introduce…

Adrien Dufraux, Emmanuel Dupoux, Awni Hannun, Armelle Brun, Matthijs Douze

December 13, 2019

Read the Paper

Publication

December 13, 2019

COMPUTER VISION

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

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,…

David Novotny, Benjamin Graham, Jeremy Reizenstein

December 13, 2019

Download the Paper

Publication

December 12, 2019

NLP

Compositional generalization through meta sequence-to-sequence learning | Facebook AI Research

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,…

Brenden M. Lake

December 12, 2019

Read the Paper

Publication

December 11, 2019

Finding the Needle in the Haystack with Convolutions: on the benefits of architectural bias | Facebook AI Research

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…

Stéphane d'Ascoli, Levent Sagun, Joan Bruna, Giulio Biroli

December 11, 2019

Download the Paper

Publication

December 11, 2019

COMPUTER VISION

SPEECH & AUDIO

Hyper-Graph-Network Decoders for Block Codes | Facebook AI Research

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.…

Eliya Nachmani, Lior Wolf

December 11, 2019

Read the Paper

Publication

December 10, 2019

Limiting Extrapolation in Linear Approximate Value Iteration | Facebook AI Research

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…