[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"detail-sidebar-cat-0-en-105":3,"doc-seo-273405-105":59,"doc-detail-273405-en":130},{"code":4,"msg":5,"data":6},0,"success",[7,13,18,23,28,33,38,43,48,51,55],{"id":8,"doc_module":4,"doc_module_name":9,"category_name":10,"show_sort_weight":11,"slug":12},1,"Document","Story & Novel",90,"story-novel",{"id":14,"doc_module":4,"doc_module_name":9,"category_name":15,"show_sort_weight":16,"slug":17},2,"Literature",80,"literature",{"id":19,"doc_module":4,"doc_module_name":9,"category_name":20,"show_sort_weight":21,"slug":22},4,"Exam",70,"exam",{"id":24,"doc_module":4,"doc_module_name":9,"category_name":25,"show_sort_weight":26,"slug":27},5,"Comic",60,"comic",{"id":29,"doc_module":4,"doc_module_name":9,"category_name":30,"show_sort_weight":31,"slug":32},6,"Technology",50,"technology",{"id":34,"doc_module":4,"doc_module_name":9,"category_name":35,"show_sort_weight":36,"slug":37},7,"Healthcare",40,"healthcare",{"id":39,"doc_module":4,"doc_module_name":9,"category_name":40,"show_sort_weight":41,"slug":42},8,"Research & Report",30,"research-report",{"id":44,"doc_module":4,"doc_module_name":9,"category_name":45,"show_sort_weight":46,"slug":47},9,"Religion & Spirituality",20,"religion-spirituality",{"id":46,"doc_module":4,"doc_module_name":9,"category_name":49,"show_sort_weight":46,"slug":50},"World Cup","world-cup",{"id":52,"doc_module":4,"doc_module_name":9,"category_name":53,"show_sort_weight":52,"slug":54},10,"Lifestyle","lifestyle",{"id":56,"doc_module":4,"doc_module_name":9,"category_name":57,"show_sort_weight":24,"slug":58},19,"General","general",{"code":4,"msg":60,"data":61},"ok",{"site_id":62,"language":63,"slug":64,"title":65,"keywords":66,"description":67,"schema_data":68,"social_meta":123,"head_meta":125,"extra_data":127,"updated_unix":129},105,"en","infuser-influence-guided-self-evolution-improves-reasoning","INFUSER: Influence-Guided Self-Evolution Improves Reasoning","","Self-evolution enables scalable improvement in large language model reasoning by using minimal external supervision, but prior approaches rely on highly curated or teacher-generated data, or use difficulty-based rewards that do not guarantee solver gains. INFUSER introduces iterative co-training with a Generator that drafts questions and golden answers from automatically collected documents and a Solver trained on them. The Generator is optimized using an influence score that directly targets solver improvement, trained with DuGRPO to handle noisy signals. Experiments show 20%+ relative gains on Olympiad and SuperGPQA and strong math/coding results, with ablations and extensions confirming flexibility.",{"@graph":69,"@context":122},[70,84,105],{"@type":71,"itemListElement":72},"BreadcrumbList",[73,77,79,82],{"item":74,"name":75,"@type":76,"position":8},"https://docshare.wps.com","Home","ListItem",{"item":78,"name":9,"@type":76,"position":14},"https://docshare.wps.com/document/",{"item":80,"name":40,"@type":76,"position":81},"https://docshare.wps.com/document/research-report/",3,{"item":83,"name":65,"@type":76,"position":19},"https://docshare.wps.com/document/infuser-influence-guided-self-evolution-improves-reasoning/273405/",{"url":83,"name":65,"@type":85,"image":86,"author":91,"headline":65,"publisher":94,"fileFormat":97,"inLanguage":63,"description":67,"dateModified":98,"datePublished":99,"encodingFormat":97,"isAccessibleForFree":100,"interactionStatistic":101},"DigitalDocument",{"url":87,"@type":88,"width":89,"height":90},"https://docshare.wps.com/thumbnails/infuser-influence-guided-self-evolution-improves-reasoning/273405.png","ImageObject",300,407,{"name":92,"@type":93},"Patrick","Person",{"url":74,"name":95,"@type":96},"DocShare","Organization","application/pdf","2026-10-18","2026-09-15",true,{"@type":102,"interactionType":103,"userInteractionCount":39},"InteractionCounter",{"@type":104},"ViewAction",{"@type":106,"mainEntity":107},"FAQPage",[108,114,118],{"name":109,"@type":110,"acceptedAnswer":111},"What problem does INFUSER address in self-evolution for reasoning models?","Question",{"text":112,"@type":113},"INFUSER targets the gap where existing self-evolution methods either require heavily curated/teacher data or use difficulty heuristics that may not improve the solver. It seeks document-grounded generator outputs that genuinely improve the current solver while co-evolving together.","Answer",{"name":115,"@type":110,"acceptedAnswer":116},"How does INFUSER train its Generator and Solver together?",{"text":117,"@type":113},"The Generator drafts questions and reference answers from automatically collected unstructured documents, while the Solver trains on these outputs using standard correctness rewards. The Generator receives an influence score that measures whether its proposed questions improve the Solver on the target distribution.",{"name":119,"@type":110,"acceptedAnswer":120},"Why introduce DuGRPO instead of standard GRPO for generator training?",{"text":121,"@type":113},"The influence score is continuous and noisy, which is poorly served by standard GRPO. DuGRPO uses a dual-normalized variant designed to better handle this noisy influence signal during generator optimization.","https://schema.org",{"og:url":83,"og:type":124,"og:title":65,"og:site_name":95,"og:description":67},"article",{"robots":126,"canonical":83},"index,follow",{"doc_id":128,"site_id":62},273405,1790020116,{"code":4,"msg":5,"data":131},{"doc_id":128,"user_id":132,"nickname":92,"user_avatar":133,"doc_module":4,"category_id":39,"category_name":40,"doc_title":65,"doc_description":67,"doc_content":134,"file_id":135,"file_url":136,"file_type":137,"file_size":138,"view_count":39,"is_deleted":4,"is_public":8,"is_downloadable":8,"audit_status":8,"page_count":139,"language":140,"language_code":63,"site_id":62,"html_lang":63,"table_of_contents":141,"faqs":142,"seo_title":143,"seo_description":67,"update_tm":144,"read_time":145,"preview_page_count":4},549758146520,"https://ap-avatar.wpscdn.com/avatar/80002397d8c0411e94?_k=1775819394049821470","arXiv : 2606 .09052v4 [cs .LG] 2 1 Aug 2026  \nINFUSER: Influence-Guided Self-Evolution Improves  \nReasoning  \nSiyu Chen 1 Miao Lu2 Beining Wu3 Heejune Sheen 1 Fengzhuo Zhang 1 Shuangning Li3 Zhiyuan Li4 Jose Blanchet2 Tianhao Wang5 Zhuoran Yang 1  \n1Yale University 2Stanford University 3University of Chicago  \n4Toyota Technological Institute at Chicago 5University of California, San Diego  \nAbstract  \nSelf-evolution offers a scalable path to stronger reasoning: a pretrained language model improves itself with only minimal external supervision. Yet existing methods either depend on extensively curated or teacher-generated training data, or, when the generator runs unsupervised, reward it by a difficulty heuristic that need not improve the solver. We introduce INFUSER, an iterative co-training framework with two co-evolving roles: a Generator that drafts questions and reference golden answers from a pool of unstructured, automatically collected documents, and a Solver that improves by training on them. The solver is trained with standard correctness rewards against the generator-provided answers, while the generator is rewarded by an optimizer-aware influence score that measures whether each proposed question would actually improve the solver on the target distribution. Because this continuous, noisy influence score is poorly served by standard GRPO, we propose DuGRPO, a dual-normalized variant of GRPO, for generator training. Together, these turn the document pool into an adaptive curriculum that favors questions useful to the current solver, not just hard ones. On Qwen3-8B-Base, INFUSER outperforms strong self-evolution baselines with over 20% relative improvement on Olympiad and SuperGPQA benchmarks, and an 8B INFUSER co-evolving generator outperforms a frozen 32B thinking generator on math and coding. Ablations confirm each design choice is necessary, and two extensions, applying INFUSER to an instruction-finetuned anchor and augmenting it with rule-verifiable RLVR data, further demonstrate the flexibility and generalizability of the framework. Code is available at [https://github.com/FFishy-git/INFUSER](https://github.com/FFishy-git/INFUSER).  \nRelative Gain over Base (%)  \n30  \n25  \n20  \n15  \n10  \n5  \n0  \nMATH500 OlympiadBench (Math)  \nMMLU-Pro SuperGPQA  \nValidation Set Accuracy  \n36  \n34  \n32  \n30  \n28  \nFixed generator  INFUSER  \nQwen3-8B-Base  \n40  \n38  \n36  \n34  \n0 50 100 Training Iteration  \nFigure 1: INFUSER on Qwen3 base anchors. Left: relative accuracy gain over Qwen3-8BBase on four headline benchmarks for each self-evolution method. Right: validation-set accuracy curves over training iterations for INFUSER versus a fixed-generator baseline with matching hyperparameters on Qwen3-4B-Base (left subpanel) and Qwen3-8B-Base (right subpanel); curves are averaged over 3 random seeds.  \nAuthor emails: [siyu.chen.sc3226@yale.edu](siyu.chen.sc3226@yale.edu), [miaolu@stanford.edu](miaolu@stanford.edu), [beiningw@uchicago.edu](beiningw@uchicago.edu),  \n[heejune.sheen@yale.edu](heejune.sheen@yale.edu), [fengzhuo.zhang@yale.edu](fengzhuo.zhang@yale.edu), [shuangning.li@chicagobooth.edu](shuangning.li@chicagobooth.edu),  \n[zhiyuanli@ttic.edu](zhiyuanli@ttic.edu), [jblanche@stanford.edu](jblanche@stanford.edu), [tianhaowang@ucsd.edu](tianhaowang@ucsd.edu), [zhuoran.yang@yale.edu](zhuoran.yang@yale.edu).  \n1 Introduction  \nReinforcement learning with verifiable rewards (RLVR) underlies much of the recent progress in reasoning for large language models (Guo et al., 2025; Kimi Team et al., 2025; Shao et al., 2024; Yu et al., 2025a; Zhang et al., 2025), but its scalability is bottlenecked by the supply of high-quality, verifiable training data, which is costly to produce in both research and industry settings. Selfevolution offers a path beyond this bottleneck: a generator proposes high-quality training data with itself or from unstructured documents, and a solver trains on that data. The whole improvement loop runs without an exter","cbCaihLGcNSAvshE","https://ap.wps.com/l/cbCaihLGcNSAvshE","pdf",5072127,72,"English","# Abstract\n# 1 Introduction\n## Reinforcement learning with verifiable rewards (RLVR)\n## Self-evolution loop and grounding\n## Generator training objective and influence-based optimization","[{\"question\":\"What problem does INFUSER address in self-evolution for reasoning models?\",\"answer\":\"INFUSER targets the gap where existing self-evolution methods either require heavily curated/teacher data or use difficulty heuristics that may not improve the solver. It seeks document-grounded generator outputs that genuinely improve the current solver while co-evolving together.\"},{\"question\":\"How does INFUSER train its Generator and Solver together?\",\"answer\":\"The Generator drafts questions and reference answers from automatically collected unstructured documents, while the Solver trains on these outputs using standard correctness rewards. The Generator receives an influence score that measures whether its proposed questions improve the Solver on the target distribution.\"},{\"question\":\"Why introduce DuGRPO instead of standard GRPO for generator training?\",\"answer\":\"The influence score is continuous and noisy, which is poorly served by standard GRPO. DuGRPO uses a dual-normalized variant designed to better handle this noisy influence signal during generator optimization.\"}]","INFUSER: Influence-Guided Self-Evolution Improves Reasoning | PDF",1789471264,181]