Inverse Reinforcement Learning Through Structured Classification
Abstract
This paper adresses the inverse reinforcement learning (IRL) problem, that is inferring a reward for which a demonstrated expert behavior is optimal. We introduce a new algorithm, SCIRL, whose principle is to use the so-called feature expectation of the expert as the parameterization of the score function of a multi-class classifier. This approach produces a reward function for which the expert policy is provably near-optimal. Contrary to most of existing IRL algorithms, SCIRL does not require solving the direct RL problem. Moreover, with an appropriate heuristic, it can succeed with only trajectories sampled according to the expert behavior. This is illustrated on a car driving simulator.
Cite
Text
Klein et al. "Inverse Reinforcement Learning Through Structured Classification." Neural Information Processing Systems, 2012.Markdown
[Klein et al. "Inverse Reinforcement Learning Through Structured Classification." Neural Information Processing Systems, 2012.](https://mlanthology.org/neurips/2012/klein2012neurips-inverse/)BibTeX
@inproceedings{klein2012neurips-inverse,
title = {{Inverse Reinforcement Learning Through Structured Classification}},
author = {Klein, Edouard and Geist, Matthieu and Piot, Bilal and Pietquin, Olivier},
booktitle = {Neural Information Processing Systems},
year = {2012},
pages = {1007-1015},
url = {https://mlanthology.org/neurips/2012/klein2012neurips-inverse/}
}