Leveraging exploration in off-policy algorithms via normalizing flows

Bogdan Mazoure, Thang Doan, Audrey Durand, Joelle Pineau, R Devon Hjelm
; Proceedings of the Conference on Robot Learning, PMLR 100:430-444, 2020.

Abstract

The ability to discover approximately optimal policies in domains with sparse rewards is crucial to applying reinforcement learning (RL) in many real-world scenarios. Approaches such as neural density models and continuous exploration (e.g., Go-Explore) have been proposed to maintain the high exploration rate necessary to find high performing and generalizable policies. Soft actor-critic (SAC) is another method for improving exploration that aims to combine efficient learning via off-policy updates, while maximizing the policy entropy. In this work, we extend SAC to a richer class of probability distributions (e.g., multimodal) through normalizing flows (NF) and show that this significantly improves performance by accelerating discovery of good policies while using much smaller policy representations. Our approach, which we call SAC-NF, is a simple, efficient, easy-to-implement modification and improvement to SAC on continuous control baselines such as MuJoCo and PyBullet Roboschool domains. Finally, SAC-NF does this while being significantly parameter efficient, using as few as 5.5% the parameters for an equivalent SAC model.

Cite this Paper


BibTeX
@InProceedings{pmlr-v100-mazoure20a, title = {Leveraging exploration in off-policy algorithms via normalizing flows}, author = {Mazoure, Bogdan and Doan, Thang and Durand, Audrey and Pineau, Joelle and Hjelm, R Devon}, booktitle = {Proceedings of the Conference on Robot Learning}, pages = {430--444}, year = {2020}, editor = {Leslie Pack Kaelbling and Danica Kragic and Komei Sugiura}, volume = {100}, series = {Proceedings of Machine Learning Research}, address = {}, month = {30 Oct--01 Nov}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v100/mazoure20a/mazoure20a.pdf}, url = {http://proceedings.mlr.press/v100/mazoure20a.html}, abstract = {The ability to discover approximately optimal policies in domains with sparse rewards is crucial to applying reinforcement learning (RL) in many real-world scenarios. Approaches such as neural density models and continuous exploration (e.g., Go-Explore) have been proposed to maintain the high exploration rate necessary to find high performing and generalizable policies. Soft actor-critic (SAC) is another method for improving exploration that aims to combine efficient learning via off-policy updates, while maximizing the policy entropy. In this work, we extend SAC to a richer class of probability distributions (e.g., multimodal) through normalizing flows (NF) and show that this significantly improves performance by accelerating discovery of good policies while using much smaller policy representations. Our approach, which we call SAC-NF, is a simple, efficient, easy-to-implement modification and improvement to SAC on continuous control baselines such as MuJoCo and PyBullet Roboschool domains. Finally, SAC-NF does this while being significantly parameter efficient, using as few as 5.5% the parameters for an equivalent SAC model.} }
Endnote
%0 Conference Paper %T Leveraging exploration in off-policy algorithms via normalizing flows %A Bogdan Mazoure %A Thang Doan %A Audrey Durand %A Joelle Pineau %A R Devon Hjelm %B Proceedings of the Conference on Robot Learning %C Proceedings of Machine Learning Research %D 2020 %E Leslie Pack Kaelbling %E Danica Kragic %E Komei Sugiura %F pmlr-v100-mazoure20a %I PMLR %J Proceedings of Machine Learning Research %P 430--444 %U http://proceedings.mlr.press %V 100 %W PMLR %X The ability to discover approximately optimal policies in domains with sparse rewards is crucial to applying reinforcement learning (RL) in many real-world scenarios. Approaches such as neural density models and continuous exploration (e.g., Go-Explore) have been proposed to maintain the high exploration rate necessary to find high performing and generalizable policies. Soft actor-critic (SAC) is another method for improving exploration that aims to combine efficient learning via off-policy updates, while maximizing the policy entropy. In this work, we extend SAC to a richer class of probability distributions (e.g., multimodal) through normalizing flows (NF) and show that this significantly improves performance by accelerating discovery of good policies while using much smaller policy representations. Our approach, which we call SAC-NF, is a simple, efficient, easy-to-implement modification and improvement to SAC on continuous control baselines such as MuJoCo and PyBullet Roboschool domains. Finally, SAC-NF does this while being significantly parameter efficient, using as few as 5.5% the parameters for an equivalent SAC model.
APA
Mazoure, B., Doan, T., Durand, A., Pineau, J. & Hjelm, R.D.. (2020). Leveraging exploration in off-policy algorithms via normalizing flows. Proceedings of the Conference on Robot Learning, in PMLR 100:430-444

Related Material