Adam Gleave

Adam’s research focuses on reward specification for reinforcement learning. Most recently, Adam developed the EPIC distance to compare reward functions. Previously, he has worked on reward modeling techniques with a number of collaborators, including inverse reinforcement learning (IRL) from vision and multi-task IRL. Adam is also interested in the robustness of reinforcement learning, and has demonstrated the existence of adversarial policies in multi-agent environments: policies that cause an opponent to fail despite behaving seemingly randomly.

Prior to joining Berkeley, Adam did a M.Phil. with Christian Steinruecken and Zoubin Ghahramani in the Machine Learning Group at the University of Cambridge. You can learn more about Adam’s work at his website and by following him on Twitter.