{"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/better-safe-than-sorry-evidence-accumulation","title":"Better Safe than Sorry: Evidence Accumulation Allows for Safe Reinforcement Learning","arxiv_id":"1809.09147","date":"2018-09-24","proceeding":null,"authors":["Akshat Agarwal","Abhinau Kumar V","Kyle Dunovan","Erik Peterson","Timothy Verstynen","Katia Sycara"],"abstract":"In the real world, agents often have to operate in situations with incomplete\ninformation, limited sensing capabilities, and inherently stochastic\nenvironments, making individual observations incomplete and unreliable.\nMoreover, in many situations it is preferable to delay a decision rather than\nrun the risk of making a bad decision. In such situations it is necessary to\naggregate information before taking an action; however, most state of the art\nreinforcement learning (RL) algorithms are biased towards taking actions\n\\textit{at every time step}, even if the agent is not particularly confident in\nits chosen action. This lack of caution can lead the agent to make critical\nmistakes, regardless of prior experience and acclimation to the environment.\nMotivated by theories of dynamic resolution of uncertainty during decision\nmaking in biological brains, we propose a simple accumulator module which\naccumulates evidence in favor of each possible decision, encodes uncertainty as\na dynamic competition between actions, and acts on the environment only when it\nis sufficiently confident in the chosen action. The agent makes no decision by\ndefault, and the burden of proof to make a decision falls on the policy to\naccrue evidence strongly in favor of a single decision. Our results show that\nthis accumulator module achieves near-optimal performance on a simple guessing\ngame, far outperforming deep recurrent networks using traditional, forced\naction selection policies.","url_abs":"http://arxiv.org/abs/1809.09147v1","url_pdf":"http://arxiv.org/pdf/1809.09147v1.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":"better-safe-than-sorry-evidence-accumulation","repo_url":"https://github.com/susumuota/gym-modeestimation","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"decision-making","task_name":"Decision Making"},{"task_slug":"reinforcement-learning","task_name":"Reinforcement Learning"},{"task_slug":"reinforcement-learning-1","task_name":"Reinforcement Learning (RL)"},{"task_slug":"safe-reinforcement-learning","task_name":"Safe Reinforcement Learning"},{"task_slug":"reinforcement-learning-2","task_name":"reinforcement-learning"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}