A Reduction of Imitation Learning and Structured Prediction to No-Regret Online Learning - Robotics Institute Carnegie Mellon University

A Reduction of Imitation Learning and Structured Prediction to No-Regret Online Learning

Conference Paper, Proceedings of 14th International Conference on Artificial Intelligence and Statistics (AISTATS '11), pp. 627 - 635, April, 2011

Abstract

Sequential prediction problems such as imitation learning, where future observations depend on previous predictions (actions), violate the common i.i.d. assumptions made in statistical learning. This leads to poor performance in theory and often in practice. Some recent approaches provide stronger guarantees in this setting, but remain somewhat unsatisfactory as they train either non-stationary or stochastic policies and require a large number of iterations. In this paper, we propose a new iterative algorithm, which trains a stationary deterministic policy, that can be seen as a no regret algorithm in an online learning setting. We show that any such no regret algorithm, combined with additional reduction assumptions, must find a policy with good performance under the distribution of observations it induces in such sequential settings. We demonstrate that this new approach outperforms previous approaches on two challenging imitation learning problems and a benchmark sequence labeling problem.

BibTeX

@conference{Ross-2011-7242,
author = {Stephane Ross and Geoffrey Gordon and J. Andrew (Drew) Bagnell},
title = {A Reduction of Imitation Learning and Structured Prediction to No-Regret Online Learning},
booktitle = {Proceedings of 14th International Conference on Artificial Intelligence and Statistics (AISTATS '11)},
year = {2011},
month = {April},
pages = {627 - 635},
keywords = {Imitation Learning, Structured Prediction, Reduction, Online Learning},
}