{"id":"https://openalex.org/W7164175407","doi":"https://doi.org/10.48550/arxiv.2606.10616","title":"Learning What to Remember: Observability-Safe Memory Retention via Constrained Optimization for Long-Horizon Language Agents","display_name":"Learning What to Remember: Observability-Safe Memory Retention via Constrained Optimization for Long-Horizon Language Agents","publication_year":2026,"publication_date":"2026-06-09","ids":{"openalex":"https://openalex.org/W7164175407","doi":"https://doi.org/10.48550/arxiv.2606.10616"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2606.10616","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.10616","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2606.10616","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5138346210","display_name":"Qingcan Kang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Kang, Qingcan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5138359027","display_name":"Liu Mingyang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Mingyang, Liu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5138301607","display_name":"Shixiong Kai","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Kai, Shixiong","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137772926","display_name":"Kaichao Liang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liang, Kaichao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5138293480","display_name":"Tao Zhong","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhong, Tao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5138322957","display_name":"Mingxuan Yuan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yuan, Mingxuan","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":null,"has_fulltext":false,"cited_by_count":0,"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/T10028","display_name":"Topic Modeling","score":0.3138999938964844,"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/T10028","display_name":"Topic Modeling","score":0.3138999938964844,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.16769999265670776,"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/T11714","display_name":"Multimodal Machine Learning Applications","score":0.08649999648332596,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"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/observability","display_name":"Observability","score":0.8346999883651733},{"id":"https://openalex.org/keywords/heuristic","display_name":"Heuristic","score":0.54339998960495},{"id":"https://openalex.org/keywords/robustness","display_name":"Robustness (evolution)","score":0.5418000221252441},{"id":"https://openalex.org/keywords/memory-model","display_name":"Memory model","score":0.4713999927043915},{"id":"https://openalex.org/keywords/context","display_name":"Context (archaeology)","score":0.4269999861717224},{"id":"https://openalex.org/keywords/constrained-optimization","display_name":"Constrained optimization","score":0.4180000126361847},{"id":"https://openalex.org/keywords/boosting","display_name":"Boosting (machine learning)","score":0.40959998965263367},{"id":"https://openalex.org/keywords/range","display_name":"Range (aeronautics)","score":0.39010000228881836}],"concepts":[{"id":"https://openalex.org/C36299963","wikidata":"https://www.wikidata.org/wiki/Q1369844","display_name":"Observability","level":2,"score":0.8346999883651733},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7473000288009644},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5523999929428101},{"id":"https://openalex.org/C173801870","wikidata":"https://www.wikidata.org/wiki/Q201413","display_name":"Heuristic","level":2,"score":0.54339998960495},{"id":"https://openalex.org/C63479239","wikidata":"https://www.wikidata.org/wiki/Q7353546","display_name":"Robustness (evolution)","level":3,"score":0.5418000221252441},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.4733000099658966},{"id":"https://openalex.org/C12186640","wikidata":"https://www.wikidata.org/wiki/Q6815743","display_name":"Memory model","level":3,"score":0.4713999927043915},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.4269999861717224},{"id":"https://openalex.org/C55660270","wikidata":"https://www.wikidata.org/wiki/Q5164377","display_name":"Constrained optimization","level":2,"score":0.4180000126361847},{"id":"https://openalex.org/C46686674","wikidata":"https://www.wikidata.org/wiki/Q466303","display_name":"Boosting (machine learning)","level":2,"score":0.40959998965263367},{"id":"https://openalex.org/C204323151","wikidata":"https://www.wikidata.org/wiki/Q905424","display_name":"Range (aeronautics)","level":2,"score":0.39010000228881836},{"id":"https://openalex.org/C76679254","wikidata":"https://www.wikidata.org/wiki/Q5165163","display_name":"Context-dependent memory","level":4,"score":0.36570000648498535},{"id":"https://openalex.org/C106189395","wikidata":"https://www.wikidata.org/wiki/Q176789","display_name":"Markov decision process","level":3,"score":0.36469998955726624},{"id":"https://openalex.org/C137836250","wikidata":"https://www.wikidata.org/wiki/Q984063","display_name":"Optimization problem","level":2,"score":0.36410000920295715},{"id":"https://openalex.org/C167966045","wikidata":"https://www.wikidata.org/wiki/Q5532625","display_name":"Generative model","level":3,"score":0.34860000014305115},{"id":"https://openalex.org/C39890363","wikidata":"https://www.wikidata.org/wiki/Q36108","display_name":"Generative grammar","level":2,"score":0.34459999203681946},{"id":"https://openalex.org/C2989514635","wikidata":"https://www.wikidata.org/wiki/Q5164377","display_name":"Constrained optimization problem","level":3,"score":0.32019999623298645},{"id":"https://openalex.org/C21200559","wikidata":"https://www.wikidata.org/wiki/Q7451068","display_name":"Sensitivity (control systems)","level":2,"score":0.31470000743865967},{"id":"https://openalex.org/C12725497","wikidata":"https://www.wikidata.org/wiki/Q810247","display_name":"Baseline (sea)","level":2,"score":0.30880001187324524},{"id":"https://openalex.org/C22367795","wikidata":"https://www.wikidata.org/wiki/Q7625208","display_name":"Structured prediction","level":2,"score":0.2624000012874603}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2606.10616","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.10616","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2606.10616","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.10616","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Long-horizon":[0],"language":[1],"agents":[2],"accumulate":[3],"observations,":[4],"reasoning":[5],"traces,":[6],"and":[7,26,53,59,105,128,140,150,158,171,173,210,221],"retrieved":[8],"facts":[9],"exceeding":[10],"context":[11],"windows,":[12],"making":[13,69],"memory":[14,41,229],"retention":[15,23,42],"a":[16,44,94,99,118,124,225],"fundamental":[17],"resource-allocation":[18],"problem.":[19],"Existing":[20],"systems":[21],"treat":[22],"as":[24,43,123,224],"local":[25],"do":[27],"not":[28],"model":[29],"long-term":[30],"consequences":[31],"under":[32,79,143,163],"observability":[33],"constraints.":[34,146],"To":[35,82],"fill":[36],"this":[37,64],"gap,":[38],"we":[39,86],"formulate":[40],"constrained":[45,218],"stochastic":[46,219],"optimization":[47,186,220],"with":[48,117],"budget":[49],"feasibility,":[50],"evidence":[51,111,116,136],"utility,":[52],"delayed":[54],"costs":[55],"including":[56],"miss,":[57],"reacquisition,":[58],"stale":[60],"penalties.":[61],"We":[62],"show":[63,152],"multi-step":[65],"problem":[66],"is":[67,187],"NP-hard,":[68],"exact":[70],"solution":[71],"intractable.":[72],"Moreover,":[73],"deployment":[74],"decisions":[75],"must":[76],"be":[77],"made":[78],"partial":[80],"observability.":[81],"address":[83],"these":[84],"challenges,":[85],"propose":[87],"OSL-MR":[88,108,153,195],"(Observability-Safe":[89],"Learning":[90],"for":[91,228],"Memory":[92],"Retention),":[93],"learning-augmented":[95],"framework":[96],"that":[97,121],"enforces":[98],"strict":[100],"separation":[101],"between":[102],"online-observable":[103],"features":[104],"offline-available":[106],"supervision.":[107],"combines":[109],"an":[110,129],"learner":[112],"trained":[113],"from":[114,137],"realized":[115],"Mixed-Score":[119,167],"heuristic":[120,160],"serves":[122],"deployable":[125,142],"online-safe":[126],"baseline":[127],"inductive":[130],"prior.":[131],"The":[132,166],"policy":[133],"learns":[134],"query-conditioned":[135],"interaction":[138],"data":[139],"remains":[141],"the":[144,200,204,207],"same":[145],"Experiments":[147],"on":[148],"LoCoMo":[149],"LongMemEval":[151],"outperforms":[154],"recency-based,":[155],"Generative":[156],"Agents-style,":[157],"other":[159],"baselines,":[161],"especially":[162],"tight":[164],"budgets.":[165],"prior":[168],"improves":[169],"precision":[170],"recall,":[172],"sensitivity":[174],"analysis":[175],"shows":[176],"robustness":[177],"across":[178],"cost":[179],"settings.":[180],"On":[181],"small":[182],"solvable":[183],"instances,":[184],"single-step":[185],"insufficient":[188],"to":[189,199],"anticipate":[190],"future":[191],"demand":[192],"shifts,":[193],"while":[194],"stays":[196],"significantly":[197],"closer":[198],"dynamic-programming":[201],"optimum,":[202],"confirming":[203],"necessity":[205],"of":[206],"sequential":[208],"formulation":[209],"reinforcing":[211],"our":[212],"learning-guided":[213],"approximation.":[214],"These":[215],"results":[216],"establish":[217],"optimization-guided":[222],"learning":[223],"principled":[226],"foundation":[227],"management":[230],"in":[231],"long-horizon":[232],"agents.":[233]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-06-11T00:00:00"}