InfoQ Homepage Facebook Content on InfoQ
-
Facebook Open-Sources Computer Vision Model Multiscale Vision Transformers
Facebook AI Research (FAIR) recently open-sourced Multiscale Vision Transformers (MViT), a deep-learning model for computer vision based on the Transformer architecture. MViT contains several internal resolution-reduction stages and outperforms other Transformer vision models while requiring less compute power, achieving new state-of-the-art accuracy on several benchmarks.
-
Facebook Superpack Leverages Code Analysis for Android App Compression
In a recent article, Facebook described its novel technique for Android app compression, Superpack, which combines compiler analysis with data compression. While not yet available for everyone, Facebook is hoping to open source it.
-
ZippyDB: the Architecture of Facebook’s Strongly Consistent Key-Value Store
Facebook Engineering recently published how it built its general-purpose key-value store, known as ZippyDB. ZippyDB is Facebook's biggest key-value store, which has been in production for more than six years. It offers flexibility to applications in terms of tunable durability, consistency, availability, and latency guarantees.
-
PyTorch 1.9 Release Includes Mobile, Scientific Computing, and Distributed Training Updates
PyTorch, Facebook's open-source deep-learning framework, announced the release of version 1.9 which includes improvements for scientific computing, mobile support, and distributed training. Overall, the new release contains more than 3,400 commits since the 1.8 release.
-
Facebook Open-Sources BlenderBot 2.0 Chatbot
Facebook AI Research (FAIR) open-sourced BlenderBot 2.0, an AI chatbot that has long-term memory and can use internet searches for supplemental conversational context. The new model outperforms version 1.0, the previous state-of-the-art chatbot, achieving 55% improvement in use of previous conversations, according to human evaluators.
-
WhatsApp Adopts the Signal Protocol for Secure Multi-Device Communication
WhatsApp is testing its new architecture aimed to enable true multi-device message synchronization while preserving end-to-end cryptographic security. To this aim, WhatsApp is adopting the Signal protocol.
-
Facebook Open-Sources Expire-Span Method for Scaling Transformer AI
Facebook AI Research (FAIR) open-sourced Expire-Span, a deep-learning technique that learns which items in an input sequence should be remembered, reducing the memory and computation requirements for AI. FAIR showed that Transformer models that incorporate Expire-Span can scale to sequences of tens of thousands of items with improved performance compared to previous models.
-
Why the Most Resilient Companies Want More Incidents
According to John Egan, the incident management process is meant to be a cycle of not just the response, but also the account of root cause and the updating of internal processes and practices across the industry. Lowering the barrier to reporting incidents, holding effective incident review meetings using blameless postmortems, and giving everyone access to postmortems is what he advises.
-
Facebook Compression Algorithm Zstandard 1.5 Improves Performance
Facebook open sourced Zstandard almost six years ago with the aim of outperforming Zlib in both speed and efficiency. Zstandard 1.5 improves compression speed at intermediate compression levels, compression ratio at higher levels, and brings faster decompression speed.
-
Relay Hooks Released, Improves User Experience with Data Prefetching
Robert Balicki and Juan Tejada, software engineers at Facebook, recently released Relay Hooks, a set of new APIs for fetching and managing GraphQL data. Relay Hooks have been battle-tested on the Facebook.com rewrite, and are the recommended way to use Relay at Facebook.
-
Facebook Announces ZionEX Platform for Training AI Models with 12 Trillion Parameters
A team of scientists at Facebook AI Research (FAIR) announced a system for training deep-learning recommendation models (DLRM) using PyTorch on a custom-built AI hardware platform, ZionEX. Using this system, the team trained models with up to 12T parameters and achieved nearly an order-of-magnitude speedup in training time compared to other systems.
-
Facebook Open-Sources AI Model to Predict COVID-19 Patient Outcomes
A team from Facebook AI Research (FAIR) and New York University (NYU) School of Medicine has developed deep-learning models that use chest X-rays to predict COVID-19 patient prognosis. In a comparison study, the models outperformed human radiologists, and could be used to help hospitals predict the demand for supplemental oxygen or intensive care.
-
Thrift for Haskell Aims to Eliminate Bugs from RPC Code
Originally created at Facebook and now part of Apache, Thrift is an interface definition language and binary communication protocol aimed to enable efficient RPC at scale across services written in multiple languages. Facebook has recently open sourced hsthrift, which makes it possible to use Thrift in Haskell projects and take advantage of its dependent types to eliminate bugs in production.
-
Facebook Open-Sources Multilingual Speech Recognition Deep-Learning Model
Facebook AI Research (FAIR) open-sourced Cross-Lingual Speech Recognition (XSLR), a multilingual speech recognition AI model. XSLR is trained on 53 languages and outperforms existing systems when evaluated on common benchmarks.
-
Facebook Open-Sources Game Playing AI ReBeL
Facebook AI Research published a paper on Recursive Belief-based Learning (ReBeL), their new AI for playing imperfect-information games that can defeat top human players in poker. The algorithm combines reinforcement learning with state-space search and converges to a Nash equilibrium for any two-player zero-sum game. Code for training the algorithm to play Liar's Dice has been open-sourced.