Off-policy Model-based Learning under Unknown Factored Dynamics

Assaf Hallak, Francois Schnitzler, Timothy Mann, Shie Mannor
Proceedings of the 32nd International Conference on Machine Learning, PMLR 37:711-719, 2015.

Abstract

Off-policy learning in dynamic decision problems is essential for providing strong evidence that a new policy is better than the one in use. But how can we prove superiority without testing the new policy? To answer this question, we introduce the G-SCOPE algorithm that evaluates a new policy based on data generated by the existing policy. Our algorithm is both computationally and sample efficient because it greedily learns to exploit factored structure in the dynamics of the environment. We present a finite sample analysis of our approach and show through experiments that the algorithm scales well on high-dimensional problems with few samples.

Cite this Paper


BibTeX
@InProceedings{pmlr-v37-hallak15, title = {Off-policy Model-based Learning under Unknown Factored Dynamics}, author = {Hallak, Assaf and Schnitzler, Francois and Mann, Timothy and Mannor, Shie}, booktitle = {Proceedings of the 32nd International Conference on Machine Learning}, pages = {711--719}, year = {2015}, editor = {Bach, Francis and Blei, David}, volume = {37}, series = {Proceedings of Machine Learning Research}, address = {Lille, France}, month = {07--09 Jul}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v37/hallak15.pdf}, url = {https://proceedings.mlr.press/v37/hallak15.html}, abstract = {Off-policy learning in dynamic decision problems is essential for providing strong evidence that a new policy is better than the one in use. But how can we prove superiority without testing the new policy? To answer this question, we introduce the G-SCOPE algorithm that evaluates a new policy based on data generated by the existing policy. Our algorithm is both computationally and sample efficient because it greedily learns to exploit factored structure in the dynamics of the environment. We present a finite sample analysis of our approach and show through experiments that the algorithm scales well on high-dimensional problems with few samples.} }
Endnote
%0 Conference Paper %T Off-policy Model-based Learning under Unknown Factored Dynamics %A Assaf Hallak %A Francois Schnitzler %A Timothy Mann %A Shie Mannor %B Proceedings of the 32nd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2015 %E Francis Bach %E David Blei %F pmlr-v37-hallak15 %I PMLR %P 711--719 %U https://proceedings.mlr.press/v37/hallak15.html %V 37 %X Off-policy learning in dynamic decision problems is essential for providing strong evidence that a new policy is better than the one in use. But how can we prove superiority without testing the new policy? To answer this question, we introduce the G-SCOPE algorithm that evaluates a new policy based on data generated by the existing policy. Our algorithm is both computationally and sample efficient because it greedily learns to exploit factored structure in the dynamics of the environment. We present a finite sample analysis of our approach and show through experiments that the algorithm scales well on high-dimensional problems with few samples.
RIS
TY - CPAPER TI - Off-policy Model-based Learning under Unknown Factored Dynamics AU - Assaf Hallak AU - Francois Schnitzler AU - Timothy Mann AU - Shie Mannor BT - Proceedings of the 32nd International Conference on Machine Learning DA - 2015/06/01 ED - Francis Bach ED - David Blei ID - pmlr-v37-hallak15 PB - PMLR DP - Proceedings of Machine Learning Research VL - 37 SP - 711 EP - 719 L1 - http://proceedings.mlr.press/v37/hallak15.pdf UR - https://proceedings.mlr.press/v37/hallak15.html AB - Off-policy learning in dynamic decision problems is essential for providing strong evidence that a new policy is better than the one in use. But how can we prove superiority without testing the new policy? To answer this question, we introduce the G-SCOPE algorithm that evaluates a new policy based on data generated by the existing policy. Our algorithm is both computationally and sample efficient because it greedily learns to exploit factored structure in the dynamics of the environment. We present a finite sample analysis of our approach and show through experiments that the algorithm scales well on high-dimensional problems with few samples. ER -
APA
Hallak, A., Schnitzler, F., Mann, T. & Mannor, S.. (2015). Off-policy Model-based Learning under Unknown Factored Dynamics. Proceedings of the 32nd International Conference on Machine Learning, in Proceedings of Machine Learning Research 37:711-719 Available from https://proceedings.mlr.press/v37/hallak15.html.

Related Material