Michael Rabbat — AMI Labs

Biography

Michael Rabbat is a Canadian researcher and technology executive whose career spans academic research, industrial AI development, and company building.

He earned his Ph.D. from the University of Wisconsin–Madison, where he developed expertise in statistical signal processing and distributed systems. Following his doctoral training, he joined McGill University in Montreal as a faculty member, where he served for more than eleven years. During his tenure at McGill, he contributed to research in machine learning, optimization, and signal processing, building a scholarly record that established him as a recognized voice in the Canadian and international machine learning communities.

Rabbat subsequently joined Meta's Fundamental AI Research (FAIR) laboratory, eventually rising to the position of Director of FAIR Montreal. In that role, he oversaw research efforts at one of the world's most prominent industrial AI research centers, working alongside leading scientists on foundational questions in machine learning. His work at FAIR became closely aligned with the emerging field of self-supervised learning and the development of systems capable of learning rich representations of the world from unlabeled data — research priorities strongly associated with FAIR's broader scientific agenda under Chief AI Scientist Yann LeCun.

Career History

2018–2024

Meta AI (FAIR)

Research Director, Montreal

2007–2018

McGill University

Associate Professor, ECE

Key Papers

Dinov2: Learning robust visual features without supervision

The recent breakthroughs in natural language processing for model pretraining on large quantities of data have opened the way for similar foundation models in computer vision. These models could greatly simplify the use of images in any system by producing all-purpose visual features, i.e., features that work across image distributions and tasks without finetuning. This work shows that existing pretraining methods, especially self-supervised methods, can produce such features if trained on enough curated data from diverse sources. We revisit existing approaches and combine different techniques to scale our pretraining in terms of data and model size. Most of the technical contributions aim at accelerating and stabilizing the training at scale. In terms of data, we propose an automatic pipeline to build a dedicated, diverse, and curated image dataset instead of uncurated data, as typically done in the self-supervised literature. In terms of models, we train a ViT model (Dosovitskiy et al., 2020) with 1B parameters and distill it into a series of smaller models that surpass the best available all-purpose features, OpenCLIP (Ilharco et al., 2021) on most of the benchmarks at image and pixel levels.

2023 · arXivLabs

10,448 citations