Timezone: »

Optimizing for the Future in Non-Stationary MDPs
Yash Chandak · Georgios Theocharous · Shiv Shankar · Martha White · Sridhar Mahadevan · Philip Thomas

Tue Jul 14 07:00 AM -- 07:45 AM & Tue Jul 14 06:00 PM -- 06:45 PM (PDT) @

Most reinforcement learning methods are based upon the key assumption that the transition dynamics and reward functions are fixed, that is, the underlying Markov decision process is stationary. However, in many real-world applications, this assumption is violated, and using existing algorithms may result in a performance lag. To proactively search for a good future policy, we present a policy gradient algorithm that maximizes a forecast of future performance. This forecast is obtained by fitting a curve to the counter-factual estimates of policy performance over time, without explicitly modeling the underlying non-stationarity. The resulting algorithm amounts to a non-uniform reweighting of past data, and we observe that minimizing performance over some of the data from past episodes can be beneficial when searching for a policy that maximizes future performance. We show that our algorithm, called Prognosticator, is more robust to non-stationarity than two online adaptation techniques, on three simulated problems motivated by real-world applications.

Author Information

Yash Chandak (University of Massachusetts Amherst)
Georgios Theocharous (Adobe Research)
Shiv Shankar (University of Massachusetts)
Martha White (University of Alberta)
Sridhar Mahadevan (Adobe Research)
Philip Thomas (University of Massachusetts Amherst)

More from the Same Authors