Sonia Joseph

I am a PhD candidate at McGill University and Mila, supervised by Blake Richards. My research focuses on mechanistic interpretability and world models, particularly in vision and video. I am interested in understanding how neural networks learn and organize representations of the physical world.

During my PhD, I was a Visiting Researcher at Meta FAIR, where I worked on interpretability of video world models. More recently, I co-founded World Mechanics, building transparent and interpretable models of the physical world.