Deprecated: The each() function is deprecated. This message will be suppressed on further calls in /home/zhenxiangba/zhenxiangba.com/public_html/phproxy-improved-master/index.php on line 456 Paper page - Experiential Reinforcement Learning
https://x.com/taiwei_shi/status/2023758117746733265\n","updatedAt":"2026-02-17T19:11:29.494Z","author":{"_id":"6039478ab3ecf716b1a5fd4d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6039478ab3ecf716b1a5fd4d/_Thy4E7taiSYBLKxEKJbT.jpeg","fullname":"taesiri","name":"taesiri","type":"user","isPro":true,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":235,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.8630950450897217},"editors":["taesiri"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/6039478ab3ecf716b1a5fd4d/_Thy4E7taiSYBLKxEKJbT.jpeg"],"reactions":[],"isReport":false}},{"id":"699518280569bcb55ab31b15","author":{"_id":"63d3e0e8ff1384ce6c5dd17d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg","fullname":"Librarian Bot (Bot)","name":"librarian-bot","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":317,"isUserFollowing":false},"createdAt":"2026-02-18T01:38:48.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"This is an automated message from the [Librarian Bot](https://huggingface.co/librarian-bots). I found the following papers similar to this paper. \n\nThe following papers were recommended by the Semantic Scholar API \n\n* [MulFeRL: Enhancing Reinforcement Learning with Verbal Feedback in a Multi-turn Loop](https://huggingface.co/papers/2601.22900) (2026)\n* [R3L: Reflect-then-Retry Reinforcement Learning with Language-Guided Exploration, Pivotal Credit, and Positive Amplification](https://huggingface.co/papers/2601.03715) (2026)\n* [KEPO: Knowledge-Enhanced Preference Optimization for Reinforcement Learning with Reasoning](https://huggingface.co/papers/2602.00400) (2026)\n* [Reinforcement Learning via Self-Distillation](https://huggingface.co/papers/2601.20802) (2026)\n* [CLEANER: Self-Purified Trajectories Boost Agentic Reinforcement Learning](https://huggingface.co/papers/2601.15141) (2026)\n* [Internalizing Meta-Experience into Memory for Guided Reinforcement Learning in Large Language Models](https://huggingface.co/papers/2602.10224) (2026)\n* [Training Multi-Turn Search Agent via Contrastive Dynamic Branch Sampling](https://huggingface.co/papers/2602.03719) (2026)\n\n\n Please give a thumbs up to this comment if you found it helpful!\n\n If you want recommendations for any Paper on Hugging Face checkout [this](https://huggingface.co/spaces/librarian-bots/recommend_similar_papers) Space\n\n You can directly ask Librarian Bot for paper recommendations by tagging it in a comment: `@librarian-bot recommend`","html":"
This is an automated message from the Librarian Bot. I found the following papers similar to this paper.
\n
The following papers were recommended by the Semantic Scholar API
Please give a thumbs up to this comment if you found it helpful!
\n
If you want recommendations for any Paper on Hugging Face checkout this Space
\n
You can directly ask Librarian Bot for paper recommendations by tagging it in a comment: \n\n@librarian-bot\n\t recommend
\n","updatedAt":"2026-02-18T01:38:48.767Z","author":{"_id":"63d3e0e8ff1384ce6c5dd17d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg","fullname":"Librarian Bot (Bot)","name":"librarian-bot","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":317,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.7503438591957092},"editors":["librarian-bot"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg"],"reactions":[],"isReport":false}},{"id":"69952e7dccda857d9f4e820c","author":{"_id":"69155b43b42f1f5fc2b06781","avatarUrl":"/avatars/d879c2075fbca19dd675094774019dfa.svg","fullname":"Anubhavasssi","name":"Anubhavassi899","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false},"createdAt":"2026-02-18T03:14:05.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"That's an awesome achievement and an alternative for model based RL. But i see it has some limitations like Higher GPU time, Larger activation memory (longer context windows), Increased training latency and depends on the reflection it generates. But it is still a better way to perform RL. ","html":"
That's an awesome achievement and an alternative for model based RL. But i see it has some limitations like Higher GPU time, Larger activation memory (longer context windows), Increased training latency and depends on the reflection it generates. But it is still a better way to perform RL.
\n","updatedAt":"2026-02-18T03:14:05.011Z","author":{"_id":"69155b43b42f1f5fc2b06781","avatarUrl":"/avatars/d879c2075fbca19dd675094774019dfa.svg","fullname":"Anubhavasssi","name":"Anubhavassi899","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.9346696734428406},"editors":["Anubhavassi899"],"editorAvatarUrls":["/avatars/d879c2075fbca19dd675094774019dfa.svg"],"reactions":[],"isReport":false}},{"id":"6995d69937dbde94da9a1c44","author":{"_id":"65243980050781c16f234f1f","avatarUrl":"/avatars/743a009681d5d554c27e04300db9f267.svg","fullname":"Avi","name":"avahal","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":3,"isUserFollowing":false},"createdAt":"2026-02-18T15:11:21.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"arXivLens breakdown of this paper 👉 https://arxivlens.com/PaperView/Details/experiential-reinforcement-learning-616-549a3863\n- Executive Summary\n- Detailed Breakdown\n- Practical Applications","html":"
\n","updatedAt":"2026-02-18T15:11:21.244Z","author":{"_id":"65243980050781c16f234f1f","avatarUrl":"/avatars/743a009681d5d554c27e04300db9f267.svg","fullname":"Avi","name":"avahal","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":3,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.7690709233283997},"editors":["avahal"],"editorAvatarUrls":["/avatars/743a009681d5d554c27e04300db9f267.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2602.13949","authors":[{"_id":"69941b5e50fb2c0be4783de6","user":{"_id":"62e1b3cb3eb0730f621a83f6","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1658958764563-noauth.jpeg","isPro":false,"fullname":"Taiwei Shi","user":"MaksimSTW","type":"user"},"name":"Taiwei Shi","status":"claimed_verified","statusLastChangedAt":"2026-02-17T15:08:55.504Z","hidden":false},{"_id":"69941b5e50fb2c0be4783de7","user":{"_id":"607d11d688160e14e4e2e4cf","avatarUrl":"/avatars/d9dff689bb705bcd40b76da1326560eb.svg","isPro":false,"fullname":"Sihao Chen","user":"sihaochen","type":"user"},"name":"Sihao Chen","status":"claimed_verified","statusLastChangedAt":"2026-02-18T09:07:12.393Z","hidden":false},{"_id":"69941b5e50fb2c0be4783de8","name":"Bowen Jiang","hidden":false},{"_id":"69941b5e50fb2c0be4783de9","user":{"_id":"64d660308ebc40443813f014","avatarUrl":"/avatars/516bb2d2383be99794e366dfb41636b6.svg","isPro":false,"fullname":"Linxin Song","user":"linxinso","type":"user"},"name":"Linxin Song","status":"claimed_verified","statusLastChangedAt":"2026-02-17T15:08:53.295Z","hidden":false},{"_id":"69941b5e50fb2c0be4783dea","name":"Longqi Yang","hidden":false},{"_id":"69941b5e50fb2c0be4783deb","name":"Jieyu Zhao","hidden":false}],"publishedAt":"2026-02-15T01:23:48.000Z","submittedOnDailyAt":"2026-02-17T05:14:40.358Z","title":"Experiential Reinforcement Learning","submittedOnDailyBy":{"_id":"62e1b3cb3eb0730f621a83f6","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1658958764563-noauth.jpeg","isPro":false,"fullname":"Taiwei Shi","user":"MaksimSTW","type":"user"},"summary":"Reinforcement learning has become the central approach for language models (LMs) to learn from environmental reward or feedback. In practice, the environmental feedback is usually sparse and delayed. Learning from such signals is challenging, as LMs must implicitly infer how observed failures should translate into behavioral changes for future iterations. We introduce Experiential Reinforcement Learning (ERL), a training paradigm that embeds an explicit experience-reflection-consolidation loop into the reinforcement learning process. Given a task, the model generates an initial attempt, receives environmental feedback, and produces a reflection that guides a refined second attempt, whose success is reinforced and internalized into the base policy. This process converts feedback into structured behavioral revision, improving exploration and stabilizing optimization while preserving gains at deployment without additional inference cost. Across sparse-reward control environments and agentic reasoning benchmarks, ERL consistently improves learning efficiency and final performance over strong reinforcement learning baselines, achieving gains of up to +81% in complex multi-step environments and up to +11% in tool-using reasoning tasks. These results suggest that integrating explicit self-reflection into policy training provides a practical mechanism for transforming feedback into durable behavioral improvement.","upvotes":55,"discussionId":"69941b5f50fb2c0be4783dec","ai_summary":"Experiential Reinforcement Learning introduces an explicit experience-reflection-consolidation loop that improves learning efficiency and performance in sparse-reward environments by enabling structured behavioral revision without additional inference costs.","ai_keywords":["reinforcement learning","environmental feedback","policy training","self-reflection","behavioral revision","exploration","optimization"],"organization":{"_id":"5e6485f787403103f9f1055e","name":"microsoft","fullname":"Microsoft","avatar":"https://cdn-uploads.huggingface.co/production/uploads/1583646260758-5e64858c87403103f9f1055d.png"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"62e1b3cb3eb0730f621a83f6","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1658958764563-noauth.jpeg","isPro":false,"fullname":"Taiwei Shi","user":"MaksimSTW","type":"user"},{"_id":"64d660308ebc40443813f014","avatarUrl":"/avatars/516bb2d2383be99794e366dfb41636b6.svg","isPro":false,"fullname":"Linxin Song","user":"linxinso","type":"user"},{"_id":"607d11d688160e14e4e2e4cf","avatarUrl":"/avatars/d9dff689bb705bcd40b76da1326560eb.svg","isPro":false,"fullname":"Sihao Chen","user":"sihaochen","type":"user"},{"_id":"64796878bb9a5693c48c9d52","avatarUrl":"/avatars/76113c49457d221fb0f3445d7b51f5dd.svg","isPro":false,"fullname":"Nattawatt Hongthong","user":"RealNattawattHongthong","type":"user"},{"_id":"620783f24e28382272337ba4","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/620783f24e28382272337ba4/zkUveQPNiDfYjgGhuFErj.jpeg","isPro":false,"fullname":"GuoLiangTang","user":"Tommy930","type":"user"},{"_id":"64ba0230c9a9894feb0cd32b","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/64ba0230c9a9894feb0cd32b/V1XWySRXAhLN4wqXZJ3Ai.jpeg","isPro":false,"fullname":"Jieyu Zhao","user":"jieyuz","type":"user"},{"_id":"684d57f26e04c265777ead3f","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/cuOj-bQqukSZreXgUJlfm.png","isPro":false,"fullname":"Joakim Lee","user":"Reinforcement4All","type":"user"},{"_id":"66935bdc5489e4f73c76bc7b","avatarUrl":"/avatars/129d1e86bbaf764b507501f4feb177db.svg","isPro":false,"fullname":"Abidoye Aanuoluwapo","user":"Aanuoluwapo65","type":"user"},{"_id":"66632a3d2dc4dff9a98c38a5","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/66632a3d2dc4dff9a98c38a5/UYk99yT6M1mZviL8d2sm_.jpeg","isPro":false,"fullname":"Yutao Mou","user":"MurrayTom","type":"user"},{"_id":"633a00248f27255b6b54ea5f","avatarUrl":"/avatars/8ad54c2d8a42093923cbdd6f15e0d7a7.svg","isPro":false,"fullname":"dfuhoiysOHSVFh82934gfjklb","user":"huba-buba","type":"user"},{"_id":"63e0a50242591dda0b9dca5c","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/63e0a50242591dda0b9dca5c/c7cBPEBWQDFYimfGnO_SI.png","isPro":false,"fullname":"Kai Zhang","user":"drogozhang","type":"user"},{"_id":"6176e8fe75f6ae94d250042a","avatarUrl":"/avatars/ed14c42b1daeb7f38d5a0956355aa077.svg","isPro":false,"fullname":"Benjamin J Schulz","user":"Burindunsmor","type":"user"}],"acceptLanguages":["*"],"dailyPaperRank":1,"organization":{"_id":"5e6485f787403103f9f1055e","name":"microsoft","fullname":"Microsoft","avatar":"https://cdn-uploads.huggingface.co/production/uploads/1583646260758-5e64858c87403103f9f1055d.png"}}">
Experiential Reinforcement Learning introduces an explicit experience-reflection-consolidation loop that improves learning efficiency and performance in sparse-reward environments by enabling structured behavioral revision without additional inference costs.
AI-generated summary
Reinforcement learning has become the central approach for language models (LMs) to learn from environmental reward or feedback. In practice, the environmental feedback is usually sparse and delayed. Learning from such signals is challenging, as LMs must implicitly infer how observed failures should translate into behavioral changes for future iterations. We introduce Experiential Reinforcement Learning (ERL), a training paradigm that embeds an explicit experience-reflection-consolidation loop into the reinforcement learning process. Given a task, the model generates an initial attempt, receives environmental feedback, and produces a reflection that guides a refined second attempt, whose success is reinforced and internalized into the base policy. This process converts feedback into structured behavioral revision, improving exploration and stabilizing optimization while preserving gains at deployment without additional inference cost. Across sparse-reward control environments and agentic reasoning benchmarks, ERL consistently improves learning efficiency and final performance over strong reinforcement learning baselines, achieving gains of up to +81% in complex multi-step environments and up to +11% in tool-using reasoning tasks. These results suggest that integrating explicit self-reflection into policy training provides a practical mechanism for transforming feedback into durable behavioral improvement.
Experiential Reinforcement Learning is a new training paradigm that shifts language model learning from simple imitation or trial and error toward structured learning from experience. Instead of just copying examples like supervised finetuning or optimizing scalar rewards like traditional reinforcement learning, it explicitly turns feedback into behavioral revision and consolidation. This reframes learning as an iterative process where experience itself becomes the primary source of improvement, marking a conceptual shift in how models adapt and learn over time.
That's an awesome achievement and an alternative for model based RL. But i see it has some limitations like Higher GPU time, Larger activation memory (longer context windows), Increased training latency and depends on the reflection it generates. But it is still a better way to perform RL.