The approach taken with I-JEPA delivers strong performance across a number of computer vision tasks and does so with greater computational efficiency compared to other widely used computer vision models.
@aiatmeta
-
I-JEPA: Self-Supervised Computer Vision Learning Through Prediction
By
–
Today we're releasing our work on I-JEPA — self-supervised computer vision that learns to understand the world by predicting it. It's the first model based on a component of @ylecun
's vision to make AI systems learn and reason like animals and humans. Details -
Meta Scales Speech Technology to 1,100+ Languages Globally
By
–
Our work on the Massively Multilingual Speech (MMS) project has scaled speech-to-text & text-to-speech to support 1,100+ languages + trained new language identification models that can identify 4,000+ languages!
— AI at Meta (@AIatMeta) 12 juin 2023
More info & access to pretrained models ⬇️Our work on the Massively Multilingual Speech (MMS) project has scaled speech-to-text & text-to-speech to support 1,100+ languages + trained new language identification models that can identify 4,000+ languages! More info & access to pretrained models
-
Next-Generation Data Center Design for Advanced AI Infrastructure
By
–
We're working on a next-generation data center design which includes higher-density racks, liquid cooling to chips and much more.
— AI at Meta (@AIatMeta) 9 juin 2023
More on how we're transforming our infrastructure for the next generation of AI ➡️ https://t.co/Dk2QzLbTIK pic.twitter.com/gf3OvtWg88We're working on a next-generation data center design which includes higher-density racks, liquid cooling to chips and much more. More on how we're transforming our infrastructure for the next generation of AI https://
bit.ly/42xCJZm -
Meta Builds Next-Generation AI Infrastructure with Custom Chips
By
–
From our first custom chip for AI models to next-generation datacenter designs optimized or AI, we're executing on an ambitious plan to build the next generation of Meta's infrastructure. More details
-
Meta’s MTIA AI Inference Accelerator: Key Problems Solved
By
–
3️⃣ Important problems that MTIA, our first-generation AI inference accelerator solves. A closer look from Mahesh Maddury, Roman Levenstein & Linda Cheng; part of the team behind this work.
— AI at Meta (@AIatMeta) 6 juin 2023
Details on Meta Training and Inference Accelerators ➡️ https://t.co/9A4I1iwNGp pic.twitter.com/aQmJ4wqwN4Important problems that MTIA, our first-generation AI inference accelerator solves. A closer look from Mahesh Maddury, Roman Levenstein & Linda Cheng; part of the team behind this work. Details on Meta Training and Inference Accelerators https://
bit.ly/42sAnv9 -
Meta AI Introduces Hiera: Faster Hierarchical Vision Transformer Model
By
–
New research from Meta AI — Hiera is an extremely simple hierarchical vision transformer that's both more accurate than previous models + significantly faster at inference and during training. Paper https://
bit.ly/45K4TU2 -
Hiera: Advanced Vision Transformer Architecture with 3.6x Speed Improvement
By
–
Hiera represents a significant advance over the vision transformer architecture. This work outperforms SOTA while being up to 3.6x faster across a range of image and video tasks — without use of domain specialized modules. Code https://
bit.ly/45OSyhq -
Meta Lattice: New Model Architecture for Enhanced Ads Performance
By
–
Meta Lattice is a new model architecture that improves Meta’s ads systems performance, AI efficiency and enables faster adaptability to the shifting market landscape.
— AI at Meta (@AIatMeta) 2 juin 2023
More on this new work ⬇️Meta Lattice is a new model architecture that improves Meta’s ads systems performance, AI efficiency and enables faster adaptability to the shifting market landscape. More on this new work
-
Meta Scalable Video Processor: First In-House ASIC for Video
By
–
Meta Scalable Processor is our first in-house designed ASIC for video processing. Alex Mastro & Junqiang Lan share more on how it's supporting some of the unique needs we face at Meta's scale. More on this work https://
bit.ly/3oIvIr1