Black-Box Policy Search with Probabilistic Programs
This work provides a novel method for policy representation in decision-making, though it appears incremental as it relates existing techniques to new applications.
The authors tackled the problem of representing policies in sequential decision problems by using probabilistic programs as black-box stochastic simulators, showing that programs can efficiently represent policies with moderate parameter counts in case studies including the Canadian traveler problem and Rock Sample.
In this work, we explore how probabilistic programs can be used to represent policies in sequential decision problems. In this formulation, a probabilistic program is a black-box stochastic simulator for both the problem domain and the agent. We relate classic policy gradient techniques to recently introduced black-box variational methods which generalize to probabilistic program inference. We present case studies in the Canadian traveler problem, Rock Sample, and a benchmark for optimal diagnosis inspired by Guess Who. Each study illustrates how programs can efficiently represent policies using moderate numbers of parameters.