{"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/a-greedy-approach-to-adapting-the-trace","title":"A Greedy Approach to Adapting the Trace Parameter for Temporal Difference Learning","arxiv_id":"1607.00446","date":"2016-07-02","proceeding":null,"authors":["Martha White","Adam White"],"abstract":"One of the main obstacles to broad application of reinforcement learning\nmethods is the parameter sensitivity of our core learning algorithms. In many\nlarge-scale applications, online computation and function approximation\nrepresent key strategies in scaling up reinforcement learning algorithms. In\nthis setting, we have effective and reasonably well understood algorithms for\nadapting the learning-rate parameter, online during learning. Such\nmeta-learning approaches can improve robustness of learning and enable\nspecialization to current task, improving learning speed. For\ntemporal-difference learning algorithms which we study here, there is yet\nanother parameter, $\\lambda$, that similarly impacts learning speed and\nstability in practice. Unfortunately, unlike the learning-rate parameter,\n$\\lambda$ parametrizes the objective function that temporal-difference methods\noptimize. Different choices of $\\lambda$ produce different fixed-point\nsolutions, and thus adapting $\\lambda$ online and characterizing the\noptimization is substantially more complex than adapting the learning-rate\nparameter. There are no meta-learning method for $\\lambda$ that can achieve (1)\nincremental updating, (2) compatibility with function approximation, and (3)\nmaintain stability of learning under both on and off-policy sampling. In this\npaper we contribute a novel objective function for optimizing $\\lambda$ as a\nfunction of state rather than time. We derive a new incremental, linear\ncomplexity $\\lambda$-adaption algorithm that does not require offline batch\nupdating or access to a model of the world, and present a suite of experiments\nillustrating the practicality of our new algorithm in three different settings.\nTaken together, our contributions represent a concrete step towards black-box\napplication of temporal-difference learning methods in real world problems.","url_abs":"http://arxiv.org/abs/1607.00446v2","url_pdf":"http://arxiv.org/pdf/1607.00446v2.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":"a-greedy-approach-to-adapting-the-trace","repo_url":"https://github.com/PwnerHarry/META","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":{"status":"ok","spdx":"MIT"}},{"paper_slug":"a-greedy-approach-to-adapting-the-trace","repo_url":"https://github.com/PwnerHarry/MTA","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":{"status":"ok","spdx":"MIT"}}],"tasks":[{"task_slug":"meta-learning","task_name":"Meta-Learning"},{"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":"speed","method_name":"SPEED"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":"https://syntology.ai/paper/1607.00446","atlas_url":"https://app.syntology.ai/?focus=1607.00446","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}