{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/safe-and-efficient-off-policy-reinforcement","title":"Safe and Efficient Off-Policy Reinforcement Learning","arxiv_id":"1606.02647","date":"2016-06-08","proceeding":"NeurIPS 2016 12","authors":["Rémi Munos","Tom Stepleton","Anna Harutyunyan","Marc G. Bellemare"],"abstract":"In this work, we take a fresh look at some old and new algorithms for\noff-policy, return-based reinforcement learning. Expressing these in a common\nform, we derive a novel algorithm, Retrace($\\lambda$), with three desired\nproperties: (1) it has low variance; (2) it safely uses samples collected from\nany behaviour policy, whatever its degree of \"off-policyness\"; and (3) it is\nefficient as it makes the best use of samples collected from near on-policy\nbehaviour policies. We analyze the contractive nature of the related operator\nunder both off-policy policy evaluation and control settings and derive online\nsample-based algorithms. We believe this is the first return-based off-policy\ncontrol algorithm converging a.s. to $Q^*$ without the GLIE assumption (Greedy\nin the Limit with Infinite Exploration). As a corollary, we prove the\nconvergence of Watkins' Q($\\lambda$), which was an open problem since 1989. We\nillustrate the benefits of Retrace($\\lambda$) on a standard suite of Atari 2600\ngames.","url_abs":"http://arxiv.org/abs/1606.02647v2","url_pdf":"http://arxiv.org/pdf/1606.02647v2.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"safe-and-efficient-off-policy-reinforcement","repo_url":"https://github.com/DanielLSM/safe-rl-tutorial","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":null},{"paper_slug":"safe-and-efficient-off-policy-reinforcement","repo_url":"https://github.com/robintyh1/icml2021-pengqlambda","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"ok","spdx":"MIT"}},{"paper_slug":"safe-and-efficient-off-policy-reinforcement","repo_url":"https://github.com/michaelnny/deep_rl_zoo","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"atari-games","task_name":"Atari Games"},{"task_slug":"reinforcement-learning","task_name":"Reinforcement Learning"},{"task_slug":"reinforcement-learning-1","task_name":"Reinforcement Learning (RL)"},{"task_slug":"reinforcement-learning-2","task_name":"reinforcement-learning"}],"methods":[{"method_slug":"retrace","method_name":"Retrace"}],"datasets_introduced":[],"methods_introduced":[{"slug":"retrace","name":"Retrace","full_name":"Retrace"}],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1606.02647","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}