https://arxiv.org/pdf/2411.18478
🌐 Project Page: https://jinyangwu.github.io/hiar-icl/

\n","updatedAt":"2024-12-02T09:04:28.367Z","author":{"_id":"6747de57f8cab58c22ec94a2","avatarUrl":"/avatars/5bae0341862fac24564781c0fa32aac5.svg","fullname":"Jinyang Wu","name":"Jinyang23","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":9,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.837812066078186},"editors":["Jinyang23"],"editorAvatarUrls":["/avatars/5bae0341862fac24564781c0fa32aac5.svg"],"reactions":[{"reaction":"👍","users":["marinaretik"],"count":1}],"isReport":false}},{"id":"674e12d7f2ad62a1ccf1a044","author":{"_id":"6486638da4cf2081f20c40ec","avatarUrl":"/avatars/0bc16a7447cd71ac18828a678313bd83.svg","fullname":"Mike Young","name":"mikelabs","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":13,"isUserFollowing":false},"createdAt":"2024-12-02T20:04:39.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"Paper summary: https://www.aimodels.fyi/papers/arxiv/beyond-examples-high-level-automated-reasoning-paradigm","html":"

Paper summary: https://www.aimodels.fyi/papers/arxiv/beyond-examples-high-level-automated-reasoning-paradigm

\n","updatedAt":"2024-12-02T20:04:39.013Z","author":{"_id":"6486638da4cf2081f20c40ec","avatarUrl":"/avatars/0bc16a7447cd71ac18828a678313bd83.svg","fullname":"Mike Young","name":"mikelabs","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":13,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.6092628836631775},"editors":["mikelabs"],"editorAvatarUrls":["/avatars/0bc16a7447cd71ac18828a678313bd83.svg"],"reactions":[],"isReport":false}},{"id":"674e601ee35fe9b6683b6e05","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":"2024-12-03T01:34:22.000Z","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* [Teaching-Inspired Integrated Prompting Framework: A Novel Approach for Enhancing Reasoning in Large Language Models](https://huggingface.co/papers/2410.08068) (2024)\n* [Reversal of Thought: Enhancing Large Language Models with Preference-Guided Reverse Reasoning Warm-up](https://huggingface.co/papers/2410.12323) (2024)\n* [Markov Chain of Thought for Efficient Mathematical Reasoning](https://huggingface.co/papers/2410.17635) (2024)\n* [AtomThink: A Slow Thinking Framework for Multimodal Mathematical Reasoning](https://huggingface.co/papers/2411.11930) (2024)\n* [Reasoning Paths Optimization: Learning to Reason and Explore From Diverse Paths](https://huggingface.co/papers/2410.10858) (2024)\n* [Improving LLM Reasoning through Scaling Inference Computation with Collaborative Verification](https://huggingface.co/papers/2410.05318) (2024)\n* [Auto-Evolve: Enhancing Large Language Model's Performance via Self-Reasoning Framework](https://huggingface.co/papers/2410.06328) (2024)\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

\n

\n
• Teaching-Inspired Integrated Prompting Framework: A Novel Approach for Enhancing Reasoning in Large Language Models (2024)
\n
• Reversal of Thought: Enhancing Large Language Models with Preference-Guided Reverse Reasoning Warm-up (2024)
\n
• Markov Chain of Thought for Efficient Mathematical Reasoning (2024)
\n
• AtomThink: A Slow Thinking Framework for Multimodal Mathematical Reasoning (2024)
\n
• Reasoning Paths Optimization: Learning to Reason and Explore From Diverse Paths (2024)
\n
• Improving LLM Reasoning through Scaling Inference Computation with Collaborative Verification (2024)
\n
• Auto-Evolve: Enhancing Large Language Model's Performance via Self-Reasoning Framework (2024)
\n

\n

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":"2024-12-03T01:34:22.262Z","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":0.7456743121147156},"editors":["librarian-bot"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg"],"reactions":[],"isReport":false}},{"id":"674edf55c180fbb7693e29d6","author":{"_id":"65a538a560cc6b04c95d750c","avatarUrl":"/avatars/60b60d6f229b397b6e19b0bb66d09e68.svg","fullname":"Antoine Frederic Grosnit","name":"AnatomyOfAFall","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":2,"isUserFollowing":false},"createdAt":"2024-12-03T10:37:09.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"I think there is a typo in equation (2), the Q value should not be divided by N(s) (you apply this when the numerator is the sum of rewards, e.g. the number of wins from a given node, but not if you have a value function).","html":"

I think there is a typo in equation (2), the Q value should not be divided by N(s) (you apply this when the numerator is the sum of rewards, e.g. the number of wins from a given node, but not if you have a value function).

\n","updatedAt":"2024-12-03T10:37:09.843Z","author":{"_id":"65a538a560cc6b04c95d750c","avatarUrl":"/avatars/60b60d6f229b397b6e19b0bb66d09e68.svg","fullname":"Antoine Frederic Grosnit","name":"AnatomyOfAFall","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":2,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.9356953501701355},"editors":["AnatomyOfAFall"],"editorAvatarUrls":["/avatars/60b60d6f229b397b6e19b0bb66d09e68.svg"],"reactions":[],"isReport":false}},{"id":"674f1dafecb7d1fab20e4f44","author":{"_id":"6747de57f8cab58c22ec94a2","avatarUrl":"/avatars/5bae0341862fac24564781c0fa32aac5.svg","fullname":"Jinyang Wu","name":"Jinyang23","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":9,"isUserFollowing":false},"createdAt":"2024-12-03T15:03:11.000Z","type":"comment","data":{"edited":true,"hidden":false,"latest":{"raw":"Thanks for your attention and pointing out this typo. In fact, in our specific implementation, the estimated reward value for each node is dynamically updated through back-propagation, which essentially involves dividing the number of wins from a given node by N(s). We'll clarify it in the next version for improved presentation.","html":"

Thanks for your attention and pointing out this typo. In fact, in our specific implementation, the estimated reward value for each node is dynamically updated through back-propagation, which essentially involves dividing the number of wins from a given node by N(s). We'll clarify it in the next version for improved presentation.

\n","updatedAt":"2024-12-04T02:27:43.371Z","author":{"_id":"6747de57f8cab58c22ec94a2","avatarUrl":"/avatars/5bae0341862fac24564781c0fa32aac5.svg","fullname":"Jinyang Wu","name":"Jinyang23","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":9,"isUserFollowing":false}},"numEdits":1,"identifiedLanguage":{"language":"en","probability":0.9429382085800171},"editors":["Jinyang23"],"editorAvatarUrls":["/avatars/5bae0341862fac24564781c0fa32aac5.svg"],"reactions":[],"isReport":false}},{"id":"67557e3484477c68c708c8c3","author":{"_id":"624e99cf1731e86cc5bf63f6","avatarUrl":"/avatars/94df01bfddd6fc8f3b8c01b24e324556.svg","fullname":"Randolphzeng","name":"Randolphzeng","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":3,"isUserFollowing":false},"createdAt":"2024-12-08T11:08:36.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"Impressive work! Just curious and for the record, any plan to opensource by any chance ? Thanks !","html":"

Impressive work! Just curious and for the record, any plan to opensource by any chance ? Thanks !

\n","updatedAt":"2024-12-08T11:08:36.062Z","author":{"_id":"624e99cf1731e86cc5bf63f6","avatarUrl":"/avatars/94df01bfddd6fc8f3b8c01b24e324556.svg","fullname":"Randolphzeng","name":"Randolphzeng","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":3,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.9079855680465698},"editors":["Randolphzeng"],"editorAvatarUrls":["/avatars/94df01bfddd6fc8f3b8c01b24e324556.svg"],"reactions":[],"isReport":false}},{"id":"6755b325c3af453d7bcbd1c9","author":{"_id":"6747de57f8cab58c22ec94a2","avatarUrl":"/avatars/5bae0341862fac24564781c0fa32aac5.svg","fullname":"Jinyang Wu","name":"Jinyang23","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":9,"isUserFollowing":false},"createdAt":"2024-12-08T14:54:29.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"Thank you for your interest in our work! We do plan to fully open source this project. Currently, we are working on supplementing and refining the experimental details. Once that is complete, we will organize the code and release it publicly.","html":"

Thank you for your interest in our work! We do plan to fully open source this project. Currently, we are working on supplementing and refining the experimental details. Once that is complete, we will organize the code and release it publicly.

\n","updatedAt":"2024-12-08T14:54:29.275Z","author":{"_id":"6747de57f8cab58c22ec94a2","avatarUrl":"/avatars/5bae0341862fac24564781c0fa32aac5.svg","fullname":"Jinyang Wu","name":"Jinyang23","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":9,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.9499484896659851},"editors":["Jinyang23"],"editorAvatarUrls":["/avatars/5bae0341862fac24564781c0fa32aac5.svg"],"reactions":[{"reaction":"🔥","users":["YuminKim"],"count":1}],"isReport":false}},{"id":"6755b5d41ab91e84e956e7ee","author":{"_id":"624e99cf1731e86cc5bf63f6","avatarUrl":"/avatars/94df01bfddd6fc8f3b8c01b24e324556.svg","fullname":"Randolphzeng","name":"Randolphzeng","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":3,"isUserFollowing":false},"createdAt":"2024-12-08T15:05:56.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"Wow. That will be really really helpful for the community! Since I saw in the comments above that you guys are preparing the next arxiv version, if you don't mind, can I suggest to add a few concrete examples for the thought cards in the appendix ? If I did not misinterepret your paper, the boost of performance lies on the smart design of using 'thought cards' to elicit/instruct the LLMs to think in a specific pattern, I am really curious about how the prompts are designed and presented. I believe that will also serve to clarify your papers a lot to some readers : )\n\n","html":"

Wow. That will be really really helpful for the community! Since I saw in the comments above that you guys are preparing the next arxiv version, if you don't mind, can I suggest to add a few concrete examples for the thought cards in the appendix ? If I did not misinterepret your paper, the boost of performance lies on the smart design of using 'thought cards' to elicit/instruct the LLMs to think in a specific pattern, I am really curious about how the prompts are designed and presented. I believe that will also serve to clarify your papers a lot to some readers : )

\n","updatedAt":"2024-12-08T15:05:56.258Z","author":{"_id":"624e99cf1731e86cc5bf63f6","avatarUrl":"/avatars/94df01bfddd6fc8f3b8c01b24e324556.svg","fullname":"Randolphzeng","name":"Randolphzeng","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":3,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.9634544849395752},"editors":["Randolphzeng"],"editorAvatarUrls":["/avatars/94df01bfddd6fc8f3b8c01b24e324556.svg"],"reactions":[],"isReport":false}},{"id":"6755c59d66bd18703ab7ac07","author":{"_id":"63fdbc250c1bbe8e29ce93f5","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1677573138583-noauth.jpeg","fullname":"ZyangLee","name":"ZyangLee","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false},"createdAt":"2024-12-08T16:13:17.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"Impressive experimental result! I still have two questions after reading this paper. \n1) How can I determine which atomic action a reasoning step in the model's output belongs to while doing MCTS? \n2) And how to prompt the model to reason guided by action-chain from selected thought card?","html":"

Impressive experimental result! I still have two questions after reading this paper.

\n

\n
1. How can I determine which atomic action a reasoning step in the model's output belongs to while doing MCTS?
\n
2. And how to prompt the model to reason guided by action-chain from selected thought card?
\n

\n","updatedAt":"2024-12-08T16:13:17.231Z","author":{"_id":"63fdbc250c1bbe8e29ce93f5","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1677573138583-noauth.jpeg","fullname":"ZyangLee","name":"ZyangLee","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.9360273480415344},"editors":["ZyangLee"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/1677573138583-noauth.jpeg"],"reactions":[],"isReport":false}},{"id":"6757026403c9069a2b88dedc","author":{"_id":"6747de57f8cab58c22ec94a2","avatarUrl":"/avatars/5bae0341862fac24564781c0fa32aac5.svg","fullname":"Jinyang Wu","name":"Jinyang23","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":9,"isUserFollowing":false},"createdAt":"2024-12-09T14:44:52.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"Insightful questions!\n\n1. For the first question, we randomly select from the entire action space. This ensures, theoretically, a deep search space, which is beneficial for constructing more comprehensive thought cards.\n2. For the second, we provide a description and functionality for each action, allowing the model to perform reasoning based on this information.","html":"

Insightful questions!

\n

\n
1. For the first question, we randomly select from the entire action space. This ensures, theoretically, a deep search space, which is beneficial for constructing more comprehensive thought cards.
\n
2. For the second, we provide a description and functionality for each action, allowing the model to perform reasoning based on this information.
\n

\n","updatedAt":"2024-12-09T14:44:52.752Z","author":{"_id":"6747de57f8cab58c22ec94a2","avatarUrl":"/avatars/5bae0341862fac24564781c0fa32aac5.svg","fullname":"Jinyang Wu","name":"Jinyang23","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":9,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.9062659740447998},"editors":["Jinyang23"],"editorAvatarUrls":["/avatars/5bae0341862fac24564781c0fa32aac5.svg"],"reactions":[],"isReport":false},"replies":[{"id":"6759014e536995a5bc547178","author":{"_id":"67492bd858e8d28424784c99","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/67492bd858e8d28424784c99/_M4_-o7-q03geacTk-3gl.jpeg","fullname":"ZyangLee","name":"LittlePoorGuy","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false},"createdAt":"2024-12-11T03:04:46.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"I remember that the MCTS realized in rStar has the restrict like One trajectory has only one action2(CoT) and action4(Re-answer subquestion)... SO you don't have restricts like above when exploring the action space? ","html":"

I remember that the MCTS realized in rStar has the restrict like One trajectory has only one action2(CoT) and action4(Re-answer subquestion)... SO you don't have restricts like above when exploring the action space?

\n","updatedAt":"2024-12-11T03:04:46.307Z","author":{"_id":"67492bd858e8d28424784c99","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/67492bd858e8d28424784c99/_M4_-o7-q03geacTk-3gl.jpeg","fullname":"ZyangLee","name":"LittlePoorGuy","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.9368134140968323},"editors":["LittlePoorGuy"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/67492bd858e8d28424784c99/_M4_-o7-q03geacTk-3gl.jpeg"],"reactions":[],"isReport":false,"parentCommentId":"6757026403c9069a2b88dedc"}},{"id":"6759014ed4d7b752340a7fb6","author":{"_id":"67492bd858e8d28424784c99","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/67492bd858e8d28424784c99/_M4_-o7-q03geacTk-3gl.jpeg","fullname":"ZyangLee","name":"LittlePoorGuy","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false},"createdAt":"2024-12-11T03:04:46.000Z","type":"comment","data":{"edited":true,"hidden":true,"hiddenBy":"","latest":{"raw":"This comment has been hidden","html":"This comment has been hidden","updatedAt":"2024-12-11T03:05:00.811Z","author":{"_id":"67492bd858e8d28424784c99","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/67492bd858e8d28424784c99/_M4_-o7-q03geacTk-3gl.jpeg","fullname":"ZyangLee","name":"LittlePoorGuy","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false}},"numEdits":0,"editors":[],"editorAvatarUrls":[],"reactions":[],"parentCommentId":"6757026403c9069a2b88dedc"}},{"id":"676e97de1eddff58354f41ac","author":{"_id":"6747de57f8cab58c22ec94a2","avatarUrl":"/avatars/5bae0341862fac24564781c0fa32aac5.svg","fullname":"Jinyang Wu","name":"Jinyang23","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":9,"isUserFollowing":false},"createdAt":"2024-12-27T12:04:46.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"Well, I just saw your comment. Thanks for the great question. Actually, during the initial construction of thought cards, we don’t impose such restrictions on the tree exploration. This is meant to create more diverse reasoning patterns for subsequent inference.","html":"

Well, I just saw your comment. Thanks for the great question. Actually, during the initial construction of thought cards, we don’t impose such restrictions on the tree exploration. This is meant to create more diverse reasoning patterns for subsequent inference.

\n","updatedAt":"2024-12-27T12:04:46.866Z","author":{"_id":"6747de57f8cab58c22ec94a2","avatarUrl":"/avatars/5bae0341862fac24564781c0fa32aac5.svg","fullname":"Jinyang Wu","name":"Jinyang23","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":9,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.9443731307983398},"editors":["Jinyang23"],"editorAvatarUrls":["/avatars/5bae0341862fac24564781c0fa32aac5.svg"],"reactions":[],"isReport":false,"parentCommentId":"6757026403c9069a2b88dedc"}}]}],"primaryEmailConfirmed":false,"paper":{"id":"2411.18478","authors":[{"_id":"674b4af16421c587617a966f","user":{"_id":"6747de57f8cab58c22ec94a2","avatarUrl":"/avatars/5bae0341862fac24564781c0fa32aac5.svg","isPro":false,"fullname":"Jinyang Wu","user":"Jinyang23","type":"user"},"name":"Jinyang Wu","status":"claimed_verified","statusLastChangedAt":"2024-12-02T08:50:22.252Z","hidden":false},{"_id":"674b4af16421c587617a9670","user":{"_id":"660d13b85e00095e45ee28e0","avatarUrl":"/avatars/8f06c01edc2a791266feadc775acb901.svg","isPro":false,"fullname":"FengMingkuan","user":"fmk345","type":"user"},"name":"Mingkuan Feng","status":"admin_assigned","statusLastChangedAt":"2024-12-02T09:16:15.000Z","hidden":false},{"_id":"674b4af16421c587617a9671","name":"Shuai Zhang","hidden":false},{"_id":"674b4af16421c587617a9672","name":"Feihu Che","hidden":false},{"_id":"674b4af16421c587617a9673","name":"Zengqi Wen","hidden":false},{"_id":"674b4af16421c587617a9674","name":"Jianhua Tao","hidden":false}],"publishedAt":"2024-11-27T16:19:00.000Z","submittedOnDailyAt":"2024-12-02T06:34:28.357Z","title":"Beyond Examples: High-level Automated Reasoning Paradigm in In-Context\n Learning via MCTS","submittedOnDailyBy":{"_id":"6747de57f8cab58c22ec94a2","avatarUrl":"/avatars/5bae0341862fac24564781c0fa32aac5.svg","isPro":false,"fullname":"Jinyang Wu","user":"Jinyang23","type":"user"},"summary":"In-context Learning (ICL) enables large language models (LLMs) to tackle\ndownstream tasks through sophisticated prompting and high-quality\ndemonstrations. However, this traditional ICL paradigm shows limitations when\nfacing complex mathematical reasoning tasks, primarily due to its heavy\ndependence on example quality and the necessity for human intervention in\nchallenging scenarios. To address these limitations, this paper presents\nHiAR-ICL, a High-level Automated Reasoning paradigm\nin ICL that shifts focus from specific examples to abstract thinking\npatterns, extending the conventional concept of context in ICL. HiAR-ICL\nintroduces five atomic reasoning actions as fundamental components for\nconstructing chain-structured patterns. Using Monte Carlo Tree Search, we\nexplore reasoning paths and construct thought cards to guide subsequent\ninference. We then develop a cognitive complexity framework that dynamically\nmatches problems with appropriate thought cards. Experimental results\ndemonstrate HiAR-ICL's effectiveness, achieving state-of-the-art accuracy\n(79.6%) on the MATH benchmark with Qwen2.5-7B-Instruct, surpassing GPT-4o\n(76.6%) and Claude 3.5 (71.1%).","upvotes":37,"discussionId":"674b4af26421c587617a9699","projectPage":"https://jinyangwu.github.io/hiar-icl/","githubRepo":"https://github.com/jinyangwu/HiARICL","githubRepoAddedBy":"auto","ai_summary":"HiAR-ICL, a high-level automated reasoning paradigm in In-context Learning (ICL), uses atomic reasoning actions, Monte Carlo Tree Search, and thought cards to improve mathematical reasoning accuracy compared to existing models.","ai_keywords":["In-context Learning","ICL","HiAR-ICL","atomic reasoning actions","Monte Carlo Tree Search","thought cards","cognitive complexity framework","MATH benchmark","Qwen2.5-7B-Instruct","GPT-4","Claude 3.5"],"githubStars":8},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"64d98ef7a4839890b25eb78b","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/64d98ef7a4839890b25eb78b/215-CSVLl81z6CAq0ECWU.jpeg","isPro":true,"fullname":"Fangyuan Yu","user":"Ksgk-fy","type":"user"},{"_id":"6747de57f8cab58c22ec94a2","avatarUrl":"/avatars/5bae0341862fac24564781c0fa32aac5.svg","isPro":false,"fullname":"Jinyang Wu","user":"Jinyang23","type":"user"},{"_id":"64e0debc02fa032de402feb0","avatarUrl":"/avatars/6c28a13216f7ae4c047cf6ea9846f6e3.svg","isPro":false,"fullname":"Chen Li","user":"LC-Edward","type":"user"},{"_id":"674d7b8fd6652a0fa71af0b9","avatarUrl":"/avatars/3b1c9d97ffca87975e1218f5b55472a8.svg","isPro":false,"fullname":"Wang","user":"Max233","type":"user"},{"_id":"64be21d81363b5c799c2dfe4","avatarUrl":"/avatars/6c678115868380715e5f257f7390d1c4.svg","isPro":false,"fullname":"shuaizhang","user":"Shuai2023","type":"user"},{"_id":"657c600badd154d163e30266","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/657c600badd154d163e30266/ChbcXfi-S1d2tIoCM5pbP.png","isPro":false,"fullname":"xander","user":"xander2077","type":"user"},{"_id":"646def60df618b303b419323","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/646def60df618b303b419323/JLJGYen4-5M8ivsLsSk0w.jpeg","isPro":false,"fullname":"Lei Wang","user":"demolei","type":"user"},{"_id":"6177a7b990ee6c00bd04cf3d","avatarUrl":"/avatars/5c517abb4a8cc8f34b840fc84dc0c3fa.svg","isPro":false,"fullname":"zhou","user":"giere","type":"user"},{"_id":"6440e79428fb8c31ffbd8e2e","avatarUrl":"/avatars/f34218b65b11e36b256f1fddd60061a6.svg","isPro":false,"fullname":"shuai nie","user":"shuai-nie90221","type":"user"},{"_id":"6411ada9d85e465a672e28a6","avatarUrl":"/avatars/76a8bf772b11437e9bd3411979284c12.svg","isPro":false,"fullname":"XIaopengWang","user":"WWWxp","type":"user"},{"_id":"674d9fe27519b842ae38d604","avatarUrl":"/avatars/9d5b0b473e871154c30669cc0d1e121b.svg","isPro":false,"fullname":"lyj","user":"hardworkinglyj","type":"user"},{"_id":"6350c89759bfa9a85d434138","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1666238674117-6350c89759bfa9a85d434138.jpeg","isPro":false,"fullname":"Yang Lee","user":"innovation64","type":"user"}],"acceptLanguages":["*"],"dailyPaperRank":3}">