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

Stephane Ross, Geoffrey Gordon, Drew Bagnell
Proceedings of the Fourteenth International Conference on Artificial Intelligence and Statistics, PMLR 15:627-635, 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.

Cite this Paper


BibTeX
@InProceedings{pmlr-v15-ross11a, title = {A Reduction of Imitation Learning and Structured Prediction to No-Regret Online Learning}, author = {Ross, Stephane and Gordon, Geoffrey and Bagnell, Drew}, booktitle = {Proceedings of the Fourteenth International Conference on Artificial Intelligence and Statistics}, pages = {627--635}, year = {2011}, editor = {Gordon, Geoffrey and Dunson, David and Dudík, Miroslav}, volume = {15}, series = {Proceedings of Machine Learning Research}, address = {Fort Lauderdale, FL, USA}, month = {11--13 Apr}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v15/ross11a/ross11a.pdf}, url = {https://proceedings.mlr.press/v15/ross11a.html}, 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.} }
Endnote
%0 Conference Paper %T A Reduction of Imitation Learning and Structured Prediction to No-Regret Online Learning %A Stephane Ross %A Geoffrey Gordon %A Drew Bagnell %B Proceedings of the Fourteenth International Conference on Artificial Intelligence and Statistics %C Proceedings of Machine Learning Research %D 2011 %E Geoffrey Gordon %E David Dunson %E Miroslav Dudík %F pmlr-v15-ross11a %I PMLR %P 627--635 %U https://proceedings.mlr.press/v15/ross11a.html %V 15 %X 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.
RIS
TY - CPAPER TI - A Reduction of Imitation Learning and Structured Prediction to No-Regret Online Learning AU - Stephane Ross AU - Geoffrey Gordon AU - Drew Bagnell BT - Proceedings of the Fourteenth International Conference on Artificial Intelligence and Statistics DA - 2011/06/14 ED - Geoffrey Gordon ED - David Dunson ED - Miroslav Dudík ID - pmlr-v15-ross11a PB - PMLR DP - Proceedings of Machine Learning Research VL - 15 SP - 627 EP - 635 L1 - http://proceedings.mlr.press/v15/ross11a/ross11a.pdf UR - https://proceedings.mlr.press/v15/ross11a.html AB - 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. ER -
APA
Ross, S., Gordon, G. & Bagnell, D.. (2011). A Reduction of Imitation Learning and Structured Prediction to No-Regret Online Learning. Proceedings of the Fourteenth International Conference on Artificial Intelligence and Statistics, in Proceedings of Machine Learning Research 15:627-635 Available from https://proceedings.mlr.press/v15/ross11a.html.

Related Material