{"id":"https://openalex.org/W7224844626","doi":"https://doi.org/10.48550/arxiv.2610.37935","title":"SkillMem: Compact External Skill Memory for Latent Multi-Agent Reasoning","display_name":"SkillMem: Compact External Skill Memory for Latent Multi-Agent Reasoning","publication_year":2026,"publication_date":"2026-10-27","ids":{"openalex":"https://openalex.org/W7224844626","doi":"https://doi.org/10.48550/arxiv.2610.37935"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2610.37935","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.37935","pdf_url":"https://arxiv.org/pdf/2610.37935","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2610.37935","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5160211579","display_name":"Yuxuan Zhu","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I165799507","display_name":"Rensselaer Polytechnic Institute","ror":"https://ror.org/01rtyzb94","country_code":"US","type":"education","lineage":["https://openalex.org/I165799507"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Zhu, Yuxuan","raw_affiliation_strings":["Rensselaer Polytechnic Institute"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Rensselaer Polytechnic Institute","institution_ids":["https://openalex.org/I165799507"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5103056406","display_name":"Mojtaba Amiri","orcid":"https://orcid.org/0000-0001-5885-6483","observed_orcids":["https://orcid.org/0000-0001-5885-6483"]},"institutions":[{"id":"https://openalex.org/I165799507","display_name":"Rensselaer Polytechnic Institute","ror":"https://ror.org/01rtyzb94","country_code":"US","type":"education","lineage":["https://openalex.org/I165799507"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Amiri, Mohammad Mohammadi","raw_affiliation_strings":["Rensselaer Polytechnic Institute"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Rensselaer Polytechnic Institute","institution_ids":["https://openalex.org/I165799507"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5160213912","display_name":"Pin-Yu Chen","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I1341412227","display_name":"IBM (United States)","ror":"https://ror.org/05hh8d621","country_code":"US","type":"company","lineage":["https://openalex.org/I1341412227"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Chen, Pin-Yu","raw_affiliation_strings":["IBM Research"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"IBM Research","institution_ids":["https://openalex.org/I1341412227"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5104188415","display_name":"Keerthiram Murugesan","orcid":"https://orcid.org/0000-0001-6847-522X","observed_orcids":["https://orcid.org/0000-0001-6847-522X"]},"institutions":[{"id":"https://openalex.org/I1341412227","display_name":"IBM (United States)","ror":"https://ror.org/05hh8d621","country_code":"US","type":"company","lineage":["https://openalex.org/I1341412227"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Murugesan, Keerthiram","raw_affiliation_strings":["IBM Research"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"IBM Research","institution_ids":["https://openalex.org/I1341412227"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":2,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":1,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10456","display_name":"Multi-Agent Systems and Negotiation","score":0.1738000065088272,"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/T10456","display_name":"Multi-Agent Systems and Negotiation","score":0.1738000065088272,"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/T10906","display_name":"AI-based Problem Solving and Planning","score":0.13899999856948853,"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/T10028","display_name":"Topic Modeling","score":0.13089999556541443,"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/secondary-storage","display_name":"external memory","score":0.6420000195503235},{"id":"https://openalex.org/keywords/inference-efficiency","display_name":"inference efficiency","score":0.5830000042915344},{"id":"https://openalex.org/keywords/memory-storage-reduction","display_name":"memory storage reduction","score":0.5759999752044678},{"id":"https://openalex.org/keywords/qwen3","display_name":"Qwen3","score":0.5839999914169312}],"concepts":[{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.7706999778747559},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7488999962806702},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.5206000208854675},{"id":"https://openalex.org/C12186640","wikidata":"https://www.wikidata.org/wiki/Q6815743","display_name":"Memory model","level":3,"score":0.5056999921798706},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5026999711990356},{"id":"https://openalex.org/C165064840","wikidata":"https://www.wikidata.org/wiki/Q1321061","display_name":"Matching (statistics)","level":2,"score":0.4742000102996826},{"id":"https://openalex.org/C44291984","wikidata":"https://www.wikidata.org/wiki/Q1074173","display_name":"Question answering","level":2,"score":0.36500000953674316},{"id":"https://openalex.org/C170133592","wikidata":"https://www.wikidata.org/wiki/Q1806883","display_name":"Latent semantic analysis","level":2,"score":0.3598000109195709},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.35280001163482666},{"id":"https://openalex.org/C125411270","wikidata":"https://www.wikidata.org/wiki/Q18653","display_name":"Encoding (memory)","level":2,"score":0.34279999136924744},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.32659998536109924},{"id":"https://openalex.org/C51167844","wikidata":"https://www.wikidata.org/wiki/Q4422623","display_name":"Latent variable","level":2,"score":0.32109999656677246},{"id":"https://openalex.org/C82687282","wikidata":"https://www.wikidata.org/wiki/Q66221","display_name":"Auxiliary memory","level":2,"score":0.31360000371932983},{"id":"https://openalex.org/C65965080","wikidata":"https://www.wikidata.org/wiki/Q1806885","display_name":"Latent variable model","level":3,"score":0.28690001368522644},{"id":"https://openalex.org/C180747234","wikidata":"https://www.wikidata.org/wiki/Q23373","display_name":"Cognitive psychology","level":1,"score":0.2775999903678894},{"id":"https://openalex.org/C67186912","wikidata":"https://www.wikidata.org/wiki/Q367664","display_name":"Data modeling","level":2,"score":0.2775000035762787},{"id":"https://openalex.org/C159032336","wikidata":"https://www.wikidata.org/wiki/Q2488768","display_name":"Non-monotonic logic","level":2,"score":0.26159998774528503},{"id":"https://openalex.org/C107457646","wikidata":"https://www.wikidata.org/wiki/Q207434","display_name":"Human\u2013computer interaction","level":1,"score":0.2533999979496002},{"id":"https://openalex.org/C176649486","wikidata":"https://www.wikidata.org/wiki/Q2308807","display_name":"Memory management","level":3,"score":0.25209999084472656}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2610.37935","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.37935","pdf_url":"https://arxiv.org/pdf/2610.37935","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2610.37935","doi":"https://doi.org/10.48550/arxiv.2610.37935","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2610.37935","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2610.37935","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.37935","pdf_url":"https://arxiv.org/pdf/2610.37935","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7224844626.pdf","grobid_xml":"https://content.openalex.org/works/W7224844626.grobid-xml"},"referenced_works_count":18,"referenced_works":["https://openalex.org/W2794325560","https://openalex.org/W3162922479","https://openalex.org/W3174770825","https://openalex.org/W3195140297","https://openalex.org/W3210904070","https://openalex.org/W4389524473","https://openalex.org/W4405254469","https://openalex.org/W4414939355","https://openalex.org/W4416037467","https://openalex.org/W7127584080","https://openalex.org/W7127616603","https://openalex.org/W7128545450","https://openalex.org/W7140299599","https://openalex.org/W7158010084","https://openalex.org/W7163699798","https://openalex.org/W7164899787","https://openalex.org/W7169575253","https://openalex.org/W7172020556"],"related_works":["https://openalex.org/W7223092157","https://openalex.org/W7163699798","https://openalex.org/W7203712181","https://openalex.org/W7160862948","https://openalex.org/W7221804984","https://openalex.org/W7127584080","https://openalex.org/W7148945997","https://openalex.org/W7127393332","https://openalex.org/W7216179301","https://openalex.org/W7164899787"],"abstract_inverted_index":{"Latent":[0],"multi-agent":[1,62,123,156,228],"systems":[2],"reduce":[3],"inference":[4,157,210],"cost":[5],"by":[6,102,194],"keeping":[7,107],"intermediate":[8],"reasoning":[9],"and":[10,55,80,89,121,136,203],"inter-agent":[11],"communication":[12],"in":[13,98],"continuous":[14],"representations.\\":[15],"Reusable":[16],"skills,":[17,189],"however,":[18],"are":[19],"commonly":[20],"supplied":[21],"as":[22,26,49],"text":[23],"or":[24,30,37],"stored":[25],"layer-wise":[27],"Key-Value":[28],"states":[29],"skill-specific":[31,176],"model":[32,159,171],"parameters,":[33,144],"introducing":[34],"repeated":[35],"processing":[36],"substantial":[38],"per-skill":[39],"storage.\\":[40],"We":[41,84,161],"introduce":[42],"\\textbf{SkillMem},":[43],"which":[44],"represents":[45],"each":[46],"reusable":[47],"skill":[48,87,105,201],"a":[50,103,185,207],"small":[51],"external":[52],"latent":[53,61,155,227],"memory":[54,71,166],"makes":[56],"it":[57],"directly":[58],"available":[59],"during":[60],"reasoning.\\":[63],"A":[64],"role-conditioned":[65],"cross-attention":[66],"reader":[67,91],"allows":[68],"the":[69,86,96,108,163,224],"same":[70,164],"to":[72,77,142,206],"guide":[73],"different":[74],"agents":[75],"according":[76],"their":[78],"roles":[79],"current":[81],"hidden":[82],"states.\\":[83],"train":[85],"memories":[88,218],"shared":[90],"through":[92],"behavioral":[93],"distillation,":[94],"matching":[95],"change":[97],"token":[99],"predictions":[100],"induced":[101],"textual":[104,116],"while":[106,152],"backbone":[109],"frozen.\\":[110],"For":[111,184],"our":[112],"experiments,":[113],"we":[114],"obtain":[115],"skills":[117,127],"from":[118,140],"paired":[119],"successful":[120],"failed":[122],"trajectories,":[124],"although":[125],"existing":[126],"can":[128,219],"also":[129],"be":[130,220],"compiled":[131],"directly.\\":[132],"Across":[133],"six":[134],"benchmarks":[135],"Qwen3":[137],"models":[138],"ranging":[139],"4B":[141],"14B":[143],"SkillMem":[145,190],"provides":[146],"strong":[147],"gains":[148],"for":[149],"smaller":[150],"backbones":[151],"preserving":[153],"efficient":[154],"across":[158,168],"scales.\\":[160],"use":[162],"$8\\times256$":[165],"configuration":[167],"all":[169],"evaluated":[170],"scales,":[172],"requiring":[173],"9.6\\,KB":[174],"of":[175,187,226],"storage":[177,193],"per":[178],"skill,":[179],"including":[180],"its":[181],"retrieval":[182],"embedding.\\":[183],"library":[186],"100":[188],"reduces":[191],"total":[192],"more":[195],"than":[196],"$38\\times$":[197],"compared":[198],"with":[199],"alternative":[200],"representations":[202],"achieves":[204],"up":[205],"$4.23\\times$":[208],"end-to-end":[209],"speedup.\\":[211],"These":[212],"results":[213],"show":[214],"that":[215],"compact":[216],"procedural":[217],"reused":[221],"without":[222],"sacrificing":[223],"efficiency":[225],"reasoning.":[229]},"counts_by_year":[{"year":2026,"cited_by_count":1}],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2026-10-31T00:00:00"}