Lucas Caccia

My current research centers on building agents that are modular and composable. I believe that modularity can enable decentralized, continual, and collaborative model development. One successful instantiation of this is model MoErging, where different (independently trained) expert models are combined into an MoE-style architecture. To this end, I have done work both on learning better experts and routing among available experts.

Prior to that, I completed in November my Ph.D. at McGill and the Quebec Artificial Intelligence Institute (Mila), where I was advised by Joelle Pineau. My PhD thesis focused on enabling efficient and robust Continual Learning in neural networks.