Jorge Méndez

Jorge A. Méndez is an MIT-IBM Distinguished Postdoctoral Fellow whose research interests are in the field of lifelong reinforcement learning, a crucial domain of machine learning that seeks to develop agents with the capacity to accumulate knowledge as they experience their environment and reuse this knowledge to adjust rapidly to change. This ability could dramatically improve the performance of ML systems in dynamic environments, from hate-speech detection models to search-and-rescue robots working in fast-changing contexts. For his doctoral research, Jorge developed one of the first general-purpose frameworks for lifelong discovery of compositional representations, which endows agents with the ability to acquire and reuse composable knowledge and combine and adapt that knowledge at scale to solve increasingly complex tasks. As a postdoctoral fellow, he continues to explore the principles of compositionality and develop strategies that enable robots to decompose complex behaviors into reusable skills, continually adapting and improving their own performance during long-term deployment. His plans involve applying the techniques of lifelong compositional learning to large-scale and long-term robot deployments, including across multiple homes, while preserving user privacy.



content Link link