We're proud to share that our paper on the Emergence of Maps in the Memories of Blind Navigation Agents was selected as one of four outstanding papers for #ICLR 2023 — congratulations @erikwijmans
, @dhruvbatradb & team! ICYMI
@aiatmeta
-
Meta’s Paper on Blind Navigation Agents Selected for ICLR 2023
By
–
-
Casual Conversations v2: Dataset for Fair and Robust AI Evaluation
By
–
The Casual Conversations v2 dataset is a consent-driven dataset of recorded monologues to enable researchers to evaluate fairness & robustness of AI models. We hope this work helps teams across the industry in building more inclusive AI & products. More details
-
ESM Atlas Updated with 740M Metagenomic Protein Structures
By
–
The updated ESM Atlas is live now and includes 740M+ metagenomic protein structures + pre-computed ESM2 embeddings for these proteins. We can't wait to see what others in the field are able to do with this expanded resource https://
bit.ly/3YUON5n 3/3 -
ESM Atlas Expands with 121M New Protein Structures
By
–
We launched the ESM Atlas in November with 600M+ metagenomic protein structures. With updates to @emblebi
's MGnify dataset, today we're expanding this with 121M new protein structures folded in just 6 days with ESMFold on spare compute capacity. 2/3 -
Meta AI Breakthrough: Language Model Accelerates Protein Folding 60x
By
–
New in @ScienceMagazine — Meta AI researchers developed a breakthrough model for protein folding by using a large language model that can accelerate folding up to 60x — with the potential to aid work in medicine, green energy & more. More details in
-
Congratulations to PyTorch on Major New Release
By
–
Congratulations @PyTorch on this week's big new release!
-
HM3D-Sem Dataset: Free 3D Scenes for Semantic Navigation Training
By
–
The HM3D-Sem dataset is free and available to use with FAIR's Habitat simulator to train embodied agents at scale for semantic navigation.
— AI at Meta (@AIatMeta) 15 mars 2023
📦 216 high-resolution 3D scenes
🏠 3.1k+ rooms
✏️ 140k+ object annotations
👤 14k+ hours of human annotation
More details ⬇️The HM3D-Sem dataset is free and available to use with FAIR's Habitat simulator to train embodied agents at scale for semantic navigation. 216 high-resolution 3D scenes 3.1k+ rooms 140k+ object annotations 14k+ hours of human annotation More details
-
Meta AI Reduces Vision Transformer Latency with Token Merging
By
–
Research from Meta AI reduces latency of existing Vision Transformer models with no additional training. Token Merging can cut inference time in half and we expect it to unlock more use of large-scale ViT models in real-world applications. Read more https://
bit.ly/3ZJv61D -
Data2vec 2.0: 16x Faster Self-Supervised Learning for Vision, Speech, Text
By
–
Created by Meta AI researchers, Data2vec 2.0 can train self-supervised models for vision, speech and text up to 16x faster than the most popular existing algorithm for images — achieving the same accuracy. Read more & access the code
-
Meta Open-Sources Casual Conversations v2 Dataset for AI Fairness
By
–
Today we’re open-sourcing Casual Conversations v2 — a consent-driven dataset of recorded monologues that includes ten self-provided & annotated categories which will enable researchers to evaluate fairness & robustness of AI models. More details on this new dataset