{"id":"https://openalex.org/W7166867199","doi":"https://doi.org/10.18653/v1/2026.findings-acl.1010","title":"Graph-GRPO: Stabilizing Multi-Agent Topology Learning via Group Relative Policy Optimization","display_name":"Graph-GRPO: Stabilizing Multi-Agent Topology Learning via Group Relative Policy Optimization","publication_year":2026,"publication_date":"2026-01-01","ids":{"openalex":"https://openalex.org/W7166867199","doi":"https://doi.org/10.18653/v1/2026.findings-acl.1010"},"language":null,"primary_location":{"id":"doi:10.18653/v1/2026.findings-acl.1010","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-acl.1010","pdf_url":"https://aclanthology.org/2026.findings-acl.1010.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: ACL 2026","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2026.findings-acl.1010.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5087209190","display_name":"Yueyang Cang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yueyang Cang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5088962332","display_name":"Xiaoteng Zhang","orcid":"https://orcid.org/0009-0002-2271-7857"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xiaoteng Zhang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5126439664","display_name":"Erlu Zhao","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Erlu Zhao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5072090101","display_name":"Zehua Ji","orcid":"https://orcid.org/0000-0003-2861-228X"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zehua Ji","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139787817","display_name":"Yuhang Liu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yuhang Liu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139716309","display_name":"Yuchen He","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yuchen He","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139811793","display_name":"Zhiyuan Ning","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhiyuan Ning","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139824878","display_name":"Yijun Chen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chen Yijun","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5108873061","display_name":"Wenge Que","orcid":"https://orcid.org/0000-0002-1385-2726"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wenge Que","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5139800929","display_name":"Li Shi","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li Shi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":{"value":0.85586233,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"20222","last_page":"20231"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10462","display_name":"Reinforcement Learning in Robotics","score":0.4957999885082245,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T10462","display_name":"Reinforcement Learning in Robotics","score":0.4957999885082245,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T11273","display_name":"Advanced Graph Neural Networks","score":0.2883000075817108,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.024399999529123306,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/group","display_name":"Group (periodic table)","score":0.4878999888896942},{"id":"https://openalex.org/keywords/topology","display_name":"Topology (electrical circuits)","score":0.4453999996185303},{"id":"https://openalex.org/keywords/stability","display_name":"Stability (learning theory)","score":0.4027000069618225},{"id":"https://openalex.org/keywords/class","display_name":"Class (philosophy)","score":0.3305000066757202},{"id":"https://openalex.org/keywords/topology-optimization","display_name":"Topology optimization","score":0.3287000060081482},{"id":"https://openalex.org/keywords/control","display_name":"Control (management)","score":0.3147999942302704}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.4968000054359436},{"id":"https://openalex.org/C2781311116","wikidata":"https://www.wikidata.org/wiki/Q83306","display_name":"Group (periodic table)","level":2,"score":0.4878999888896942},{"id":"https://openalex.org/C184720557","wikidata":"https://www.wikidata.org/wiki/Q7825049","display_name":"Topology (electrical circuits)","level":2,"score":0.4453999996185303},{"id":"https://openalex.org/C112972136","wikidata":"https://www.wikidata.org/wiki/Q7595718","display_name":"Stability (learning theory)","level":2,"score":0.4027000069618225},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.37929999828338623},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.35749998688697815},{"id":"https://openalex.org/C2777212361","wikidata":"https://www.wikidata.org/wiki/Q5127848","display_name":"Class (philosophy)","level":2,"score":0.3305000066757202},{"id":"https://openalex.org/C189216461","wikidata":"https://www.wikidata.org/wiki/Q2443456","display_name":"Topology optimization","level":3,"score":0.3287000060081482},{"id":"https://openalex.org/C2775924081","wikidata":"https://www.wikidata.org/wiki/Q55608371","display_name":"Control (management)","level":2,"score":0.3147999942302704},{"id":"https://openalex.org/C199845137","wikidata":"https://www.wikidata.org/wiki/Q145490","display_name":"Network topology","level":2,"score":0.29010000824928284},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.27160000801086426},{"id":"https://openalex.org/C2776401178","wikidata":"https://www.wikidata.org/wiki/Q12050496","display_name":"Feature (linguistics)","level":2,"score":0.26269999146461487},{"id":"https://openalex.org/C2983025851","wikidata":"https://www.wikidata.org/wiki/Q1040098","display_name":"Group learning","level":2,"score":0.2619999945163727},{"id":"https://openalex.org/C2778112365","wikidata":"https://www.wikidata.org/wiki/Q3511065","display_name":"Sequence (biology)","level":2,"score":0.25619998574256897},{"id":"https://openalex.org/C64543145","wikidata":"https://www.wikidata.org/wiki/Q162942","display_name":"Intersection (aeronautics)","level":2,"score":0.2549000084400177}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2026.findings-acl.1010","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-acl.1010","pdf_url":"https://aclanthology.org/2026.findings-acl.1010.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: ACL 2026","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2026.findings-acl.1010","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-acl.1010","pdf_url":"https://aclanthology.org/2026.findings-acl.1010.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: ACL 2026","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7166867199.pdf","grobid_xml":"https://content.openalex.org/works/W7166867199.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Optimizing":[0],"communication":[1,103,166],"topology":[2,81,94],"is":[3],"fundamental":[4],"to":[5,23],"the":[6,48,110,121,126,133],"efficiency":[7],"and":[8,47,108,140,148,163],"effectiveness":[9],"of":[10,90,101,112],"Large":[11],"Language":[12],"Model":[13],"(LLM)-based":[14],"Multi-Agent":[15],"Systems":[16],"(MAS).While":[17],"recent":[18],"approaches":[19],"utilize":[20],"reinforcement":[21],"learning":[22,71],"dynamically":[24],"construct":[25],"task-specific":[26],"graphs,":[27],"they":[28],"typically":[29],"rely":[30],"on":[31,116,146],"single-sample":[32],"policy":[33],"gradients":[34],"with":[35],"absolute":[36],"rewards":[37,57,124],"(e.g.,":[38],"binary":[39],"correctness).This":[40],"paradigm":[41],"suffers":[42],"from":[43,136],"severe":[44],"gradient":[45],"variance":[46,139],"credit":[49,143],"assignment":[50],"problem:":[51],"simple":[52],"queries":[53,63],"yield":[54],"non-informative":[55],"positive":[56],"for":[58,105],"suboptimal":[59],"structures,":[60],"while":[61],"difficult":[62],"often":[64],"result":[65],"in":[66,95],"failures":[67],"that":[68,84,153],"provide":[69],"no":[70],"signal.To":[72],"address":[73],"these":[74],"challenges,":[75],"we":[76],"propose":[77],"Graph-GRPO,":[78],"a":[79,92,99],"novel":[80],"optimization":[82],"framework":[83],"integrates":[85],"Group":[86],"Relative":[87],"Policy":[88],"Optimization.Instead":[89],"evaluating":[91],"single":[93],"isolation,":[96],"Graph-GRPO":[97,154],"samples":[98],"group":[100],"diverse":[102],"graphs":[104],"each":[106],"query":[107],"computes":[109],"advantage":[111],"specific":[113],"edges":[114],"based":[115],"their":[117],"relative":[118],"performance":[119],"within":[120],"group.By":[122],"normalizing":[123],"across":[125],"sampled":[127],"group,":[128],"our":[129],"method":[130],"effectively":[131],"mitigates":[132],"noise":[134],"derived":[135],"task":[137],"difficulty":[138],"enables":[141],"finegrained":[142],"assignment.Extensive":[144],"experiments":[145],"reasoning":[147],"code":[149],"generation":[150],"benchmarks":[151],"demonstrate":[152],"significantly":[155],"outperforms":[156],"state-of-the-art":[157],"baselines,":[158],"achieving":[159],"superior":[160],"training":[161],"stability":[162],"identifying":[164],"critical":[165],"pathways":[167],"previously":[168],"obscured":[169],"by":[170],"reward":[171],"noise.":[172]},"counts_by_year":[],"updated_date":"2026-07-29T14:22:42.915294","created_date":"2026-07-02T00:00:00"}