ML APPLICATIONS

How Meta measures the management of its AI ecosystem

AI development ecosystems are increasingly complex and challenging to maintain. We're sharing details on our work to develop highly efficient systems to build, serve and improve AI models for production uses.

March 24, 2023

ML APPLICATIONS

How Meta measures the management of its AI ecosystem

March 24, 2023

RESEARCH

Introducing Casual Conversations v2: A more inclusive dataset to measure fairness

Casual Conversations v2

March 09, 2023

ML APPLICATIONS

MuAViC: The first audio-video speech translation benchmark

Introducing MuAViC — A new benchmark for audio-visual learning for robust speech translation. We used it to train model to translate speech in noisy, challenging settings, achieving results outperforming other leading translation models.

March 08, 2023

Blog Posts

Blog

August 31, 2022

RESEARCH

Using AI to decode speech from brain activity

Decoding speech from brain activity has mostly benefited from invasive approaches. New research from FAIR shows AI could instead make use of noninvasive brain scans.

August 31, 2022

Learn More

Blog

August 11, 2022

RESEARCH

COMPUTER VISION

Implicitron: A new modular, extensible framework for neural implicit representations in PyTorch3D

We are releasing Implicitron, an extension of PyTorch3D that enables fast prototyping of 3D reconstruction and new-view synthesis methods based on rendering of implicit representations such as radiance fields, signed distance fields, and more.

August 11, 2022

Learn More

Blog

August 05, 2022

RESEARCH

NLP

Blender BlenderBot 3: A 175B parameter, publicly available chatbot that improves its skills and safety over time

We’re announcing that Meta AI has built and released BlenderBot 3, the first 175B parameter, publicly available chatbot complete with model weights, code, datasets, and model cards. We’ve deployed it in a live interactive conversational AI demo.

August 05, 2022

Learn More

Blog

August 01, 2022

RESEARCH

COMPUTER VISION

We’re sharing a new approach for building computer vision systems that can more accurately recognize a wide range of common and uncommon objects. By using a simpler Transformer architecture together with Mask R-CNN and self-supervised pretraining, our ViTDet system outperforms previous methods.

We’re sharing a new approach for building computer vision systems that can more accurately recognize a wide range of common and uncommon objects. By using a simpler Transformer architecture together with Mask R-CNN and self-supervised pretraining, our ViTDet system outperforms previous methods.

August 01, 2022

Learn More

Blog

July 28, 2022

RESEARCH

ML APPLICATIONS

Assessing fairness of our products while protecting people’s privacy

To help Meta better understand how people across different ethnic and racial groups experience our products, we are launching a voluntary survey for people in the U.S. who use Instagram. The survey uses sophisticated encryption to prevent anyone, including us, from connecting individuals with their survey responses.

July 28, 2022

Learn More

Blog

July 26, 2022

RESEARCH

NLP

Applications of Meta’s OPT-175B: Protein design, quantum physics, translation, and more

We’ve seen an amazing response to OPT-175B, the first large language model of its kind to be made freely available to the research community. We’ve provided OPT access to hundreds of universities, industry research labs, and other entities.

July 26, 2022

Learn More

Blog

July 20, 2022

RESEARCH

COMPUTER VISION

Introducing Theseus, a library for encoding domain knowledge in end to end AI models

Meta AI is open-sourcing Theseus, a library for incorporating expert domain knowledge into AI architectures for robotics and vision models. Read the paper and get the code.

July 20, 2022

Learn More

Blog

July 14, 2022

COMPUTER VISION

Greater creative control for AI image generation

We’re showcasing an exploratory AI research concept called Make-A-Scene. This multimodal generative AI method puts creative control in the hands of people who use it by allowing them to describe and illustrate their vision through both text descriptions and freeform sketches.

July 14, 2022

Learn More

Blog

July 06, 2022

RESEARCH

NLP

200 languages within a single AI model: A breakthrough in high-quality machine translation

Meta AI has built a single AI model, NLLB-200, that is the first to translate across 200 different languages with state-of-the-art quality that has been validated through extensive evaluations for each of them.

July 06, 2022

Learn More

Blog

July 04, 2022

NLP

We’ve created a new white-box retrieval solution using the world’s vastest library — the open web — as a universal, uncurated, and unstructured source of knowledge, to solve multiple KI-NLP tasks at once.

We’ve created a new white-box retrieval solution using the world’s vastest library — the open web — as a universal, uncurated, and unstructured source of knowledge, to solve multiple KI-NLP tasks at once.