We examine the problem of learning mappings from state to state, suitable for use in a model-based reinforcement-learning setting, that simultaneously generalize to novel states and can capture stochastic transitions. We show that currently popular generative adversarial networks struggle to learn these stochastic transition models but a modification to their loss functions results in a powerful learning algorithm for this class of problems.
@article{arxiv.1710.09718,
title = {Learning Approximate Stochastic Transition Models},
author = {Yuhang Song and Christopher Grimm and Xianming Wang and Michael L. Littman},
journal= {arXiv preprint arXiv:1710.09718},
year = {2017}
}