Michael Rabbat

Also known as Mike Rabbat

Michael Rabbat is a machine learning researcher whose work runs from gossip algorithms for distributed averaging to self-supervised world models. A PhD graduate of the University of Wisconsin-Madison, he was a professor in McGill University's Department of Electrical and Computer Engineering and coauthored a paper on greedy gossip with eavesdropping in 2008. In 2017, he was among the first four researchers to open Meta's Montreal FAIR Lab, and he coauthored the fastMRI dataset and Stochastic Gradient Push in 2018, I-JEPA in 2023 and V-JEPA 2 in 2025. He is a former director of Meta's FAIR Montreal group and is Co-Founder and VP World Models at Advanced Machine Intelligence (AMI), the company Yann LeCun co-founded after leaving Meta, which raised $1.03 billion in March 2026.

Now at: Advanced Machine Intelligence and Mila
Badges: Founder

Timeline

2008–2009

Gossip algorithms at McGill

A PhD graduate of the University of Wisconsin-Madison, he is a professor in McGill University's Department of Electrical and Computer Engineering, where the Computer Networks Research Lab lists him with Mark Coates and doctoral students Deniz Ustebay and Boris Oreshkin. They publish "Rates of Convergence for Greedy Gossip with Eavesdropping" at the Allerton Conference in September 2008. Greedy gossip with eavesdropping is a randomized gossip algorithm for computing the average consensus of values held across a network, and the full paper reaches arXiv in 2009.
2017

Opening the FAIR Montreal lab

In 2017, he is one of the first four researchers to open Meta's Montreal FAIR Lab, together with Pascal Vincent, Joelle Pineau and Nicolas Ballas. The lab, led by Pineau, takes an open-science approach and sets up partnerships with CIFAR, Mila, McGill University and Université de Montréal.
2018

The fastMRI dataset

In 2018, he is one of 27 authors of "fastMRI: An Open Dataset and Benchmarks for Accelerated MRI." The dataset holds raw MR measurements and clinical MR images for training machine-learning approaches to MRI reconstruction, and it comes out of a partnership between Montreal FAIR researchers and NYU Langone Health.
2018

Stochastic Gradient Push

In 2018, he coauthors "Stochastic Gradient Push for Distributed Deep Learning" with Mahmoud Assran, Nicolas Loizou and Nicolas Ballas. The method combines the PushSum gossip algorithm with stochastic gradient updates, and the paper proves that it converges to a stationary point of smooth, non-convex objectives at the same sub-linear rate as SGD.
2023

I-JEPA for images

In January 2023, he coauthors I-JEPA with Mahmoud Assran, Yann LeCun, Nicolas Ballas and others. The non-generative method learns image representations by predicting the representations of target blocks from a single context block in the same image.
June 2025

V-JEPA 2 for video

In June 2025, he is a coauthor of V-JEPA 2, a joint-embedding predictive architecture pre-trained on over 1 million hours of internet video. A latent action-conditioned world model built on it, V-JEPA 2-AC, is applied to robotic planning.
March 2026

AMI Labs raises $1.03 billion

AMI Labs, the world-model venture Yann LeCun co-founded after leaving Meta, raises $1.03 billion in March 2026. Its team lists Rabbat as VP of world models, with Saining Xie as chief science officer and Pascale Fung as chief research and innovation officer.
As of October 2026

AMI co-founder in Montreal

He is Co-Founder and VP World Models at Advanced Machine Intelligence (AMI), based in Montreal, and an affiliate member of Mila. His research interests include efficient and robust representation learning, self-supervised learning, and model-based planning and control. He is a former director of Meta's FAIR Montreal group.