[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-86215-en":3,"doc-seo-86215-105":30,"detail-sidebar-cat-0-en-105":92},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":20,"is_deleted":4,"is_public":21,"is_downloadable":21,"audit_status":21,"page_count":22,"language":23,"language_code":24,"site_id":25,"html_lang":24,"table_of_contents":26,"faqs":27,"seo_title":13,"seo_description":14,"update_tm":28,"read_time":29},86215,1374391974564,"Clementine","https://ap-avatar.wpscdn.com/avatar/14000253aa45c000a9e?x-image-process=image/resize,m_fixed,w_180,h_180&k=1779874745381141002",8,"Research & Report","RefineEvo Planning-Guided Heuristic Evolution with Bidirectional Experience","Automatic Heuristic Design (AHD) addresses combinatorial optimization by discovering effective heuristics with reduced manual labor. While LLM-based AHD methods show promise, they often rely on fixed evolutionary operators and do not reliably accumulate or reuse historical search experience. RefineEvo converts AHD into a planning-guided, experience-driven framework with a Planner for adaptive operator scheduling and refinement, and a Reflector that distills trajectory-aware bidirectional lessons into an Experience Pool. Experiments show stronger solution quality and better token efficiency.","RefineEvo: Planning-Guided Heuristic Evolution with Bidirectional Experience  \nYang Wu * 1 2 Junran Pan * 1 2 Yifan Zhang 1 2 3 Ning Xu 1 Fanshuo Zeng 1 2 Jian Cheng 1 2 4  \narXiv :2607 . 1 1358v 1 [ cs .CL] 13 Jul 2026  \nAbstract  \nAutomatic Heuristic Design (AHD) has emerged as a transformative approach for solving combinatorial optimization problems. While recent Large Language Model (LLM)-based methods have shown promise, they predominantly rely on fixed evolutionary operators and struggle to effectively accumulate and reuse historical search experience. This paper proposes RefineEvo, a novel evolutionary framework that transforms AHD from a static trial-and-error process into a planning-guided, experience-driven system. RefineEvo introduces a Planner to dynamically schedule evolutionary operators and trigger refinement based on the current search state, and a Reflector to distill valuable lessons into a Bidirectional  \nExperience Pool containing both positive insightsand negative pitfalls. This synergistic framework enables the system to adapt its search tools to the evolving complexity of the problem and leverage trajectory-aware, situation-conditioned insights to guide generation. Experiments on several classic combinatorial optimization benchmarks demonstrate that RefineEvo consistently outperforms strong baselines. In particular, RefineEvo delivers superior solution quality while improving token efficiency, enabling more efficient and autonomous heuristic design.  \n1. Introduction  \nAutomatic Heuristic Design (AHD) has emerged as a promising avenue for automating the discovery of effective heuristics, alleviating the burden of labor-intensiv˜ e manual  \ndesign (Burke et al., 2013 ; St¨utzle & Lpez-Ibnez, 2018) .  \nThe recent integration of Large Language Models (LLMs)  \n1 C2DL, Institute of Automation, Chinese Academy of Sciences 2 School of Artificial Intelligence, University of Chinese Academy of Sciences, Beijing 3University of Chinese Academy of Sciences, Nanjing 4AIRIA. Correspondence to: Yifan Zhang \u003C[yfzhang@nlpr.ia.ac.cn](yfzhang@nlpr.ia.ac.cn) >.  \nProceedings of the 43 rd International Conference on Machine Learning, Seoul, South Korea. PMLR 306, 2026 . Copyright 2026 by the author(s) .  \nfurther accelerates this process (Romera-Paredes et al., 2024 ; Liu et al., 2024b) by enabling semantic, structureaware code transformations (Chen et al., 2021 ; Li et al., 2022 ; Achiam et al., 2023) that can generate genuinely novel algorithms. However, this capability also opens up an effectively unbounded program space (Wu et al., 2024 ; Liu et al., 2024c ; Dong et al., 2025), making it crucial to harness LLMs to find superior solutions with minimal trial-and-error and computation.  \nEvolutionary operators in LLM-based AHD are typically implemented as prompt templates, ranging from explorationoriented prompts that guide the LLM to mutate or crossover code to refinement-oriented prompts that improve existing solutions. Pioneering works (Romera-Paredes et al., 2024 ; Liu et al., 2024b) have compellingly demonstrated the effectiveness of coupling LLMs with evolutionary search by maintaining a fixed pool of pre-defined operators and applying them broadly across candidate algorithms. However, this “use-all” strategy overlooks operator heterogeneity and state-dependent utility (Di Tollo et al., 2015 ; Pei et al., 2025), so many operator applications are misaligned with the candidate’s current stability and progress. Furthermore, as the evolutionary process advances, finding valid improvements becomes increasingly challenging (Ye et al., 2025 ; Cui et al., 2025) . This necessitates the continuous refinement of the operators themselves to adapt to the evolving complexity of the search space.  \nParallel to the focus on operators, complementary efforts have been directed towards accumulating experience from past iterations (Ye et al., 2024 ; Dat et al., 2025 ; Chen et al., 2025 ; Liu et al., 2025) . These methods primarily employ","cbCaipx6aWjHUUSr","https://ap.wps.com/l/cbCaipx6aWjHUUSr","pdf",2317510,5,1,43,"English","en",105,"# Introduction\n## Background: LLM-based AHD and fixed operators\n## Gap: misaligned operator utility and harder improvements\n## Gap: outcome-only experience and missing parent-relative context\n## Proposed framework overview: RefineEvo","[{\"question\":\"What problem does RefineEvo address in LLM-based automatic heuristic design?\",\"answer\":\"It targets two issues: fixed evolutionary operators that may not fit the current search state, and experience accumulation that is outcome-based and ignores parent-to-offspring relative improvement and situation context.\"},{\"question\":\"How does RefineEvo decide which evolutionary operator to use and when to refine operators?\",\"answer\":\"A Planner dynamically schedules evolutionary operators and triggers refinement based on the current search state and population/operator statistics.\"},{\"question\":\"What is the role of bidirectional experience in RefineEvo?\",\"answer\":\"A Reflector distills structured lessons from parent-to-offspring trajectories into a Bidirectional Experience Pool, containing both useful insights and misleading pitfalls, so guidance is conditioned on situations.\"}]",1784209525,108,{"code":4,"msg":31,"data":32},"ok",{"site_id":25,"language":24,"slug":33,"title":13,"keywords":34,"description":14,"schema_data":35,"social_meta":87,"head_meta":89,"extra_data":91,"updated_unix":28},"refineevo-planning-guided-heuristic-evolution-with-bidirectional-experience","",{"@graph":36,"@context":86},[37,54,69],{"@type":38,"itemListElement":39},"BreadcrumbList",[40,44,48,51],{"item":41,"name":42,"@type":43,"position":21},"https://docshare.wps.com","Home","ListItem",{"item":45,"name":46,"@type":43,"position":47},"https://docshare.wps.com/document/","Document",2,{"item":49,"name":12,"@type":43,"position":50},"https://docshare.wps.com/document/research-report/",3,{"item":52,"name":13,"@type":43,"position":53},"https://docshare.wps.com/document/refineevo-planning-guided-heuristic-evolution-with-bidirectional-experience/86215/",4,{"url":52,"name":13,"@type":55,"author":56,"headline":13,"publisher":58,"fileFormat":61,"inLanguage":24,"description":14,"dateModified":62,"datePublished":63,"encodingFormat":61,"isAccessibleForFree":64,"interactionStatistic":65},"DigitalDocument",{"name":9,"@type":57},"Person",{"url":41,"name":59,"@type":60},"DocShare","Organization","application/pdf","2026-07-27","2026-07-16",true,{"@type":66,"interactionType":67,"userInteractionCount":20},"InteractionCounter",{"@type":68},"ViewAction",{"@type":70,"mainEntity":71},"FAQPage",[72,78,82],{"name":73,"@type":74,"acceptedAnswer":75},"What problem does RefineEvo address in LLM-based automatic heuristic design?","Question",{"text":76,"@type":77},"It targets two issues: fixed evolutionary operators that may not fit the current search state, and experience accumulation that is outcome-based and ignores parent-to-offspring relative improvement and situation context.","Answer",{"name":79,"@type":74,"acceptedAnswer":80},"How does RefineEvo decide which evolutionary operator to use and when to refine operators?",{"text":81,"@type":77},"A Planner dynamically schedules evolutionary operators and triggers refinement based on the current search state and population/operator statistics.",{"name":83,"@type":74,"acceptedAnswer":84},"What is the role of bidirectional experience in RefineEvo?",{"text":85,"@type":77},"A Reflector distills structured lessons from parent-to-offspring trajectories into a Bidirectional Experience Pool, containing both useful insights and misleading pitfalls, so guidance is conditioned on situations.","https://schema.org",{"og:url":52,"og:type":88,"og:title":13,"og:site_name":59,"og:description":14},"article",{"robots":90,"canonical":52},"index,follow",{"doc_id":7,"site_id":25},{"code":4,"msg":5,"data":93},[94,98,102,106,110,115,120,123,128,131,135],{"id":21,"doc_module":4,"doc_module_name":46,"category_name":95,"show_sort_weight":96,"slug":97},"Story & Novel",90,"story-novel",{"id":47,"doc_module":4,"doc_module_name":46,"category_name":99,"show_sort_weight":100,"slug":101},"Literature",80,"literature",{"id":53,"doc_module":4,"doc_module_name":46,"category_name":103,"show_sort_weight":104,"slug":105},"Exam",70,"exam",{"id":20,"doc_module":4,"doc_module_name":46,"category_name":107,"show_sort_weight":108,"slug":109},"Comic",60,"comic",{"id":111,"doc_module":4,"doc_module_name":46,"category_name":112,"show_sort_weight":113,"slug":114},6,"Technology",50,"technology",{"id":116,"doc_module":4,"doc_module_name":46,"category_name":117,"show_sort_weight":118,"slug":119},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":46,"category_name":12,"show_sort_weight":121,"slug":122},30,"research-report",{"id":124,"doc_module":4,"doc_module_name":46,"category_name":125,"show_sort_weight":126,"slug":127},9,"Religion & Spirituality",20,"religion-spirituality",{"id":126,"doc_module":4,"doc_module_name":46,"category_name":129,"show_sort_weight":126,"slug":130},"World Cup","world-cup",{"id":132,"doc_module":4,"doc_module_name":46,"category_name":133,"show_sort_weight":132,"slug":134},10,"Lifestyle","lifestyle",{"id":136,"doc_module":4,"doc_module_name":46,"category_name":137,"show_sort_weight":20,"slug":138},19,"General","general"]