August 12, 2015

No Regret Bound for Extreme Bandits

ArXiv PrePrint

A sensible notion of regret in the extreme bandit setting

Robert Nishihara, David Lopez-Paz, Leon Bottou
June 25, 2015

Scale-Invariant Learning and Convolutional Networks

ArXiv PrePrint

The conventional classification schemes — notably multinomial logistic regression — used in conjunction with convolutional networks (convnets) are classical in statistics, designed without consideration for the usual coupling with convnets, stochastic gradient descent, and backpropagation. In the specific application to supervised learning for convnets, a simple scale-invariant classification stage turns out to be more robust than multinomial logistic regression, appears to result in slightly lower errors on several standard test sets, has similar computational costs, and features precise control over the actual rate of learning.

Arthur Szlam, Marc'Aurelio Ranzato, Mark Tygert, Soumith Chintala, Wojciech Zaremba, Yuandong Tian
June 22, 2015

End-To-End Memory Networks

NIPS 2015

End-to-end training of Memory Networks on question answering and to language modeling

Sainbayar Sukhbaatar, Arthur Szlam, Jason Weston, Rob Fergus
June 22, 2015

Learning Longer Memory in Recurrent Neural Networks

Workshop at ICLR 2015

We describe simple extension of recurrent networks that allows them to learn longer term memory. Usefulness is demonstrated in improved performance on standard language modeling tasks.

Tomas Mikolov, Armand Joulin, Sumit Chopra, Michael Mathieu, Marc'Aurelio Ranzato
June 22, 2015

Large-Scale Simple Question Answering with Memory Networks

ArXiv PrePrint

This paper studies the impact of multitask and transfer learning for simple question answering; a setting for which the reasoning required to answer is quite easy, as long as one can retrieve the correct evidence given a question, which can be difficult in large-scale conditions.

Antoine Bordes, Jason Weston, Nicolas Usunier, Sumit Chopra
June 22, 2015

Revisiting Memory Errors in Large-Scale Production Data Centers: Analysis and Modeling of New Trends from the Field

IEEE/IFIP International Conference on Dependable Systems and Networks

In this paper, we analyze the memory errors in the entire fleet of servers at Facebook over the course of fourteen months, representing billions of device days.

Justin Meza, Qiang Wu, Sanjeev Kumar, Onur Mutlu
June 22, 2015

An Implementation of a Randomized Algorithm for Principal Component Analysis

ArXiv PrePrint

This paper carefully implements newly popular randomized algorithms for principal component analysis and benchmarks them against the classics.

Arthur Szlam, Mark Tygert, Yuval Kluger
June 22, 2015

Learning Spatiotemporal Features with 3D Convolutional Networks

ArXiv PrePrint

We propose C3D, a simple and effective approach for spatiotemporal feature using deep 3-dimensional convolutional networks (3D ConvNets) trained on a large scale supervised video dataset.

Du Tran, Lubomir Bourdev, Rob Fergus, Lorenzo Torresani, Manohar Paluri
June 22, 2015

Deep Generative Image Models using a Laplacian Pyramid of Adversarial Networks

ArXiv PrePrint

We introduce a generative parametric model capable of producing high quality samples of natural images

Emily Denton, Soumith Chintala, Arthur Szlam, Rob Fergus
June 22, 2015

Fast Convolutional Nets With fbfft: A GPU Performance Evaluation

International Conference on Learning Representations, 2015

We examine the performance profile of Convolutional Neural Network training on the current generation of NVIDIA Graphics Processing Units.

Nicolas Vasilache, Jeff Johnson, Soumith Chintala, Serkan Piantino, Yann LeCun
June 22, 2015

Inferring Algorithmic Patterns with Stack-Augmented Recurrent Nets

ArXiv PrePrint

We develop a model that overcomes certain basic limitations of popular deep learning models. We demonstrate its capabilities by learning in an unsupervised way concepts such as simple memorization and binary addition.

Armand Joulin, Tomas Mikolov
June 22, 2015

A Theoretical Argument for Complex-Valued Convolutional Networks

ArXiv PrePrint

This article provides foundations for certain convolutional networks.

Arthur Szlam, Joan Bruna, Mark Tygert, Serkan Piantino, Soumith Chintala, Yann LeCun
June 15, 2015

A Large-Scale Study of Flash Memory Failures in the Field

ACM Sigmetrics 2015

This paper presents the first large-scale study of flash-based SSD reliability in the field.

Justin Meza, Qiang Wu, Sanjeev Kumar, Onur Mutlu
June 12, 2015

Web-Scale Training for Face Identification

The IEEE Conference on Computer Vision and Pattern Recognition

We study face recognition and show that three distinct properties have surprising effects on the transferability of deep convolutional networks (CNN)

Yaniv Taigman, Ming Yang, Marc'Aurelio Ranzato, Lior Wolf
June 1, 2015

Beyond Frontal Faces: Improving Person Recognition Using Multiple Cues

International Conference on Computer Vision and Pattern Recognition

We propose a method for person recognition from arbitrary viewpoint and pose.

Ning Zhang, Manohar Paluri, Yaniv Taigman, Rob Fergus, Lubomir Bourdev
May 19, 2015

Challenges to Adopting Stronger Consistency at Scale

Workshop on Hot Topics in Operating Systems

There have been many recent advances in distributed systems that provide stronger semantics for geo-replicated data stores like those underlying Facebook. At Facebook we are excited by these lines of research, but fundamental and operational challenges currently make it infeasible to incorporate these advances into deployed systems. This paper describes some of these challenges with the hope that future advances will address them.

Philippe Ajoux, Nathan Bronson, Sanjeev Kumar, Wyatt Lloyd, Kaushik Veeraraghavan
May 18, 2015

The Lifecycles of Apps in a Social Ecosystem

Proc. WWW'15

Apps are emerging as an important form of on-line content, and they combine aspects of Web usage in interesting ways — they exhibit a rich temporal structure of user adoption and long-term engagement, and they exist in a broader social ecosystem that helps drive these patterns of adoption and engagement. It has been difficult, however, to study apps in their natural setting since this requires a simultaneous analysis of a large set of popular apps and the underlying social network they inhabit.

Isabel Kloumann, Lada Adamic, Jon Kleinberg, Shaomei Wu
May 18, 2015

Design and Analysis of Benchmarking Experiments for Distributed Internet Services

Proc. WWW

We develop statistical models of distributed Internet service performance based on data from Perflab, a production system used at Facebook which vets thousands of changes to the company’s codebase each day.

Eytan Bakshy, Eitan Frachtenberg
May 9, 2015

Exposure to Ideologically Diverse Information on Facebook


How do these online networks influence exposure to perspectives that cut across ideological lines?

Eytan Bakshy, Solomon Messing, Lada Adamic
May 6, 2015

Wormhole: Reliable Pub-Sub to Support Geo-replicated Internet Services

12th USENIX Symposium on Networked Systems Design and Implementation

Wormhole is a publish-subscribe (pub-sub) system developed for use within Facebook’s geographically replicated datacenters. It is used to reliably replicate changes among several Facebook services including TAO, Graph Search and Memcache. This paper describes the design and implementation of Wormhole as well as the operational challenges of scaling the system to support the multiple data storage systems deployed at Facebook.

Yogeshwer Sharma, Philippe Ajoux, Petchean Ang, David Callies, Abhishek Choudhary, Laurent Demailly, Thomas Fersch, Liat Atsmon, Andrzej Kotulski, Sachin Kulkarni, Sanjeev Kumar, Hu Li, Jun Li, Evgeniy Makeev, Kowshik Prakasam, Robbert van Renesse, Sabyasachi Roy, Pratyush Seth, Yee Jiun Song, Kaushik Veeraraghavan, Benjamin Wester, Peter Xie