Nan Jiang, a CSE PhD candidate, has been awarded a Rackham Predoctoral Fellowship to support his research while he completes his dissertation, which is entitled, A Theory of Model Selection in Batch Reinforcement Learning. The Rackham Predoctoral Fellowship is awarded to outstanding doctoral candidates in the final stages of their program who are unusually creative, ambitious and risk-taking.
In his dissertation, he investigates the overfitting phenomenon in batch Reinforcement Learning (RL) and provides theoretical and empirical results on how to perform model selection in the RL context. The topic is central to continuing the success of RL from simulator-defined problems (e.g., video games, Computer Go) to real-life problems (e.g., adaptive medical treatment, personalized education, dialogue systems); these problems are typically characterized by the large size of state (and action) spaces and relatively limited amount of data, where overfitting could easily occur without a careful choice of representation, and how to perform model selection to avoid overfitting is still much of an open problem. In his work, he advances the theoretical understanding of overfitting in RL, characterizes the difficulties of existing strategies for model selection (such as cross-validation), and develops new algorithms with provable guarantees.
Nan Jiang is advised by Prof. Satinder Baveja Singh.
About the Rackham Predoctoral Fellowship
The Rackham Predoctoral Fellowship supports outstanding doctoral students who have achieved candidacy and are actively working on dissertation research and writing. They seek to support students working on dissertation that are unusually creative, ambitious and risk-taking.
Posted: May 23, 2016