Products

To enable speech recognition technology for many more languages spoken around the globe, Facebook AI is releasing wav2vec Unsupervised, a new method to train models with no supervision whatsoever. Wav2vec Unsupervised rivals the performance of the best supervised systems from just a few years ago.

May 21, 2021

Learn More

Blog

May 24, 2021

Research

Dynaboard: Moving beyond accuracy to holistic model evaluation in NLP

We're introducing Dynaboard, an evaluation-as-a-service platform for conducting comprehensive evaluations of NLP models.

May 24, 2021

Learn More

Blog

May 19, 2021

Research

How AI is learning to see the bigger picture

Facebook CTO Michael Schroepfer shares an overview of how new advances in AI are helping us better enforce our Community Standards.

May 19, 2021

Learn More

Blog

May 14, 2021

Research

Teaching AI how to forget at scale

As a step toward achieving humanlike memory in machines, we’re announcing Expire-Span, a first-of-its-kind method that equips neural networks with the ability to forget at scale.

May 14, 2021

Learn More

Blog

April 30, 2021

Research

Computer Vision

DINO and PAWS: Advancing the state of the art in computer vision

Working with Inria researchers, we’ve developed a self-supervised image representation method, DINO, which produces remarkable results when trained with Vision Transformers. We are also detailing PAWS, a new method for 10x more efficient training.

April 30, 2021

Learn More

Blog

April 26, 2021

Computer Vision

Many AI systems can’t recognize both of these examples. SEER can.

Many computer vision systems can recognize the item on the left but not the one on the right. Using self-supervised learning, our new SEER model overcomes this limitation, which can help us create #computervision systems that work well for everyone around the globe.

April 26, 2021

Learn More

Blog

April 22, 2021

Research

Facebook AI and Facebook Reality Labs announce research and mentorship program with Carnegie Mellon University

Excited to announce the expansion of a long-standing collaboration with Carnegie Mellon @SCSatCMU. The Facebook Research & AI Mentorship (FRAIM) program gives PhD students at CMU a unique opportunity to perform their research in collaboration with their thesis adviser and a Facebook researcher. Learn more:

April 22, 2021

Learn More

Blog

April 16, 2021

Research

Open Source

AI predicts effective drug combinations to fight complex diseases faster

In partnership with Helmholtz Munich, we’ve built & open sourced a new AI model that will accelerate discovery of new drug combinations. It's the first single model to predict effects of combinations, dosages, timing, & interventions like gene deletion.

April 16, 2021

Learn More

Blog

April 16, 2021

Open Source

Flashlight: Fast and flexible machine learning in C++

Deep and machine learning (ML) frameworks are good at what they do, but altering their underlying code is typically difficult. Flashlight is a new open source library written entirely in C++ that enables teams to rapidly modify frameworks to fit their needs. Read more:

April 16, 2021

Learn More

Blog

April 15, 2021

ML Applications

Computer Vision

How maps built with Facebook AI can help with COVID-19 vaccine delivery

Our population density maps can be used to coordinate and improve the delivery of humanitarian aid around the world, including COVID-19 vaccinations.

April 15, 2021

Learn More

Blog

April 13, 2021

Research

Computer Vision

Q&A with Abhinav Gupta, winner of the J.K. Aggarwal Prize

Congratulations to Abhinav Gupta, on receiving this year's J.K. Aggarwal Prize. Abhinav took a moment to tell us more about the impact his work has had on computer vision and robotics, why AI algorithms should be more like babies, and where he plans to go from here. Learn more:

April 13, 2021

Learn More

Blog

April 05, 2021

Research

ML Applications

Large-scale forecasting: Self-supervised learning framework for hyperparameter tuning

Unlike existing hyperparameter tuning methods, our new self-supervised learning framework for hyperparameter tuning is not search-based. It uses time series features as inputs and produces optimal hyperparameters in 6-20x less time — without sacrificing accuracy.

April 05, 2021

Learn More

Prev Next