{"id":"https://openalex.org/W7170034571","doi":"https://doi.org/10.1109/mlise70044.2026.11607496","title":"Failure-Gated Hierarchical Memory: Preventing Memory Pollution in Long-Horizon LLM Agents via Critic-Guided Write Control","display_name":"Failure-Gated Hierarchical Memory: Preventing Memory Pollution in Long-Horizon LLM Agents via Critic-Guided Write Control","alternate_titles":[],"publication_year":2026,"publication_date":"2026-05-01","ids":{"openalex":"https://openalex.org/W7170034571","doi":"https://doi.org/10.1109/mlise70044.2026.11607496"},"language":"en","primary_location":{"id":"doi:10.1109/mlise70044.2026.11607496","doi":"https://doi.org/10.1109/mlise70044.2026.11607496","is_oa":false,"landing_page_url":"https://doi.org/10.1109/mlise70044.2026.11607496","pdf_url":null,"source":{"id":"https://openalex.org/S4363608615","display_name":"International Conference on Machine Learning and Intelligent Systems Engineering (MLISE)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/P4310319808","host_organization_name":"Institute of Electrical and Electronics Engineers","host_organization_lineage":["https://openalex.org/P4310319808"],"host_organization_lineage_names":["Institute of Electrical and Electronics Engineers"],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2026 6th International Conference on Machine Learning and Intelligent Systems Engineering (MLISE)","raw_type":"proceedings-article"},"parent_works":[],"lineage":["https://openalex.org/W7170034571"],"child_works_count":0,"type":"conference-paper","is_traditional":true,"indexed_in":["crossref"],"open_access":{"is_oa":false,"oa_status":"closed","oa_url":null,"any_repository_has_fulltext":false},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5143417502","display_name":"Qizhi Liu","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I4210148850","display_name":"National Science Library","ror":"https://ror.org/04ndjyr10","country_code":"CN","type":"archive","lineage":["https://openalex.org/I19820366","https://openalex.org/I4210148850"]}],"countries":["CN"],"is_corresponding":false,"contributor_roles":[],"raw_author_name":"Qizhi Liu","raw_affiliation_strings":["National Science Library, Beijing, China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"National Science Library, Beijing, China","institution_ids":["https://openalex.org/I4210148850"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5079264255","display_name":"M Wang","orcid":"https://orcid.org/0009-0006-2920-0454","observed_orcids":["https://orcid.org/0009-0006-2920-0454","https://orcid.org/0009-0007-8833-2884","https://orcid.org/0000-0002-9606-6308","https://orcid.org/0009-0002-9832-1920","https://orcid.org/0009-0003-1087-6911","https://orcid.org/0009-0003-1917-8941","https://orcid.org/0009-0003-4939-4653","https://orcid.org/0009-0003-6538-0809","https://orcid.org/0009-0004-4791-5037","https://orcid.org/0009-0005-0855-5638","https://orcid.org/0009-0007-4550-4017","https://orcid.org/0009-0009-3628-7721"]},"institutions":[{"id":"https://openalex.org/I4210148850","display_name":"National Science Library","ror":"https://ror.org/04ndjyr10","country_code":"CN","type":"archive","lineage":["https://openalex.org/I19820366","https://openalex.org/I4210148850"]}],"countries":["CN"],"is_corresponding":false,"contributor_roles":[],"raw_author_name":"Meng Wang","raw_affiliation_strings":["National Science Library, Beijing, China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"National Science Library, Beijing, China","institution_ids":["https://openalex.org/I4210148850"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5143425383","display_name":"Jing Xie","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I4210148850","display_name":"National Science Library","ror":"https://ror.org/04ndjyr10","country_code":"CN","type":"archive","lineage":["https://openalex.org/I19820366","https://openalex.org/I4210148850"]}],"countries":["CN"],"is_corresponding":false,"contributor_roles":[],"raw_author_name":"Jing Xie","raw_affiliation_strings":["National Science Library, Beijing, China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"National Science Library, Beijing, China","institution_ids":["https://openalex.org/I4210148850"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I4210148850"],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":false,"cited_by_count":0,"reviewed_by_count":0,"citation_normalized_percentile":{"value":0.0,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"132","last_page":"139"},"is_retracted":false,"is_pulled":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10906","display_name":"AI-based Problem Solving and Planning","score":0.8490999937057495,"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/T10906","display_name":"AI-based Problem Solving and Planning","score":0.8490999937057495,"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":0.0203000009059906,"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/T10462","display_name":"Reinforcement Learning in Robotics","score":0.017400000244379044,"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/llm-based-agents","display_name":"LLM agents","score":0.734000027179718},{"id":"https://openalex.org/keywords/hierarchical-memory","display_name":"hierarchical memory","score":0.5929999947547913},{"id":"https://openalex.org/keywords/long-term-memory","display_name":"long-term memory","score":0.8619999885559082},{"id":"https://openalex.org/keywords/episodic-buffer","display_name":"episodic buffer","score":0.7680000066757202},{"id":"https://openalex.org/keywords/retrieval-augmented-generation","display_name":"retrieval-augmented generation","score":0.8029999732971191},{"id":"https://openalex.org/keywords/alfworld","display_name":"ALFWorld","score":0.6200000047683716},{"id":"https://openalex.org/keywords/webshop","display_name":"WebShop","score":0.7179999947547913}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6068000197410583},{"id":"https://openalex.org/C2775924081","wikidata":"https://www.wikidata.org/wiki/Q55608371","display_name":"Control (management)","level":2,"score":0.508899986743927},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.3215000033378601},{"id":"https://openalex.org/C2778258037","wikidata":"https://www.wikidata.org/wiki/Q7225750","display_name":"Pollution prevention","level":2,"score":0.2775000035762787},{"id":"https://openalex.org/C17500928","wikidata":"https://www.wikidata.org/wiki/Q959968","display_name":"Control system","level":2,"score":0.272599995136261},{"id":"https://openalex.org/C2780791683","wikidata":"https://www.wikidata.org/wiki/Q846785","display_name":"Action (physics)","level":2,"score":0.2590000033378601},{"id":"https://openalex.org/C127413603","wikidata":"https://www.wikidata.org/wiki/Q11023","display_name":"Engineering","level":0,"score":0.25690001249313354},{"id":"https://openalex.org/C112930515","wikidata":"https://www.wikidata.org/wiki/Q4389547","display_name":"Risk analysis (engineering)","level":1,"score":0.2459000051021576},{"id":"https://openalex.org/C149635348","wikidata":"https://www.wikidata.org/wiki/Q193040","display_name":"Embedded system","level":1,"score":0.23469999432563782},{"id":"https://openalex.org/C155386361","wikidata":"https://www.wikidata.org/wiki/Q1649571","display_name":"Process control","level":3,"score":0.23430000245571136}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1109/mlise70044.2026.11607496","doi":"https://doi.org/10.1109/mlise70044.2026.11607496","is_oa":false,"landing_page_url":"https://doi.org/10.1109/mlise70044.2026.11607496","pdf_url":null,"source":{"id":"https://openalex.org/S4363608615","display_name":"International Conference on Machine Learning and Intelligent Systems Engineering (MLISE)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/P4310319808","host_organization_name":"Institute of Electrical and Electronics Engineers","host_organization_lineage":["https://openalex.org/P4310319808"],"host_organization_lineage_names":["Institute of Electrical and Electronics Engineers"],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2026 6th International Conference on Machine Learning and Intelligent Systems Engineering (MLISE)","raw_type":"proceedings-article"}],"best_oa_location":null,"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":21,"referenced_works":["https://openalex.org/W1878893887","https://openalex.org/W3027879771","https://openalex.org/W3041214984","https://openalex.org/W3126325318","https://openalex.org/W3168867926","https://openalex.org/W4386528753","https://openalex.org/W4387636003","https://openalex.org/W4393065402","https://openalex.org/W4393147158","https://openalex.org/W4393160747","https://openalex.org/W4402684150","https://openalex.org/W4407759090","https://openalex.org/W4412203333","https://openalex.org/W4412887782","https://openalex.org/W4415797504","https://openalex.org/W4415971869","https://openalex.org/W7127163877","https://openalex.org/W7133208539","https://openalex.org/W7133222210","https://openalex.org/W7133227044","https://openalex.org/W7166848218"],"related_works":["https://openalex.org/W7204791447","https://openalex.org/W7165924858","https://openalex.org/W7164198312","https://openalex.org/W7172054893","https://openalex.org/W7166701541","https://openalex.org/W7168197591","https://openalex.org/W7109003516","https://openalex.org/W7171559101","https://openalex.org/W7166845515","https://openalex.org/W7133724598"],"abstract_inverted_index":{"Large":[0],"language":[1],"model":[2,105],"(LLM)-based":[3],"agents":[4],"equipped":[5],"with":[6,80,153],"persistent":[7],"memory":[8,16,19,70,183,233],"can":[9],"accumulate":[10],"experience":[11],"across":[12],"tasks,":[13],"yet":[14],"uncontrolled":[15],"writes":[17,89],"introduce":[18,141],"pollution\u2014erroneous,":[20],"outdated,":[21],"or":[22,118],"over-generalized":[23],"knowledge":[24,109,156],"entries":[25,54],"that":[26,86,132,243],"silently":[27],"degrade":[28],"downstream":[29,161],"task":[30,195],"performance":[31],"through":[32],"retrievalaugmented":[33],"generation.":[34],"This":[35],"phenomenon":[36],"is":[37,247],"particularly":[38],"harmful":[39],"in":[40],"long-horizon":[41,251],"deployments":[42],"where":[43,145],"the":[44,91,135,146,154,181,214],"agent":[45,85,252],"continuously":[46],"operates":[47],"over":[48,56,204],"hundreds":[49],"of":[50,137,199,217],"episodes,":[51],"as":[52,101,227],"polluted":[53],"compound":[55],"time":[57],"via":[58],"cascading":[59],"error":[60],"propagation.":[61],"We":[62,139,220],"present":[63],"Failure-Gated":[64],"Hierarchical":[65],"Memory":[66],"(FGHM),":[67],"a":[68,81,102,128,194,228],"three-tier":[69],"architecture":[71],"(working":[72],"memory,":[73],"episodic":[74,92],"buffer,":[75],"procedural":[76,95],"longterm":[77,96],"memory)":[78],"augmented":[79],"dedicated":[82],"Failure":[83],"Critic":[84,147],"gates":[87],"all":[88],"from":[90,112],"buffer":[93],"to":[94,158],"memory.":[97],"The":[98],"Critic,":[99],"implemented":[100],"LoRA-tuned":[103],"Llama-3-8B":[104],"trained":[106],"on":[107,166],"hard-negative":[108],"candidates":[110,126],"derived":[111],"failure":[113,151],"trajectories,":[114],"issues":[115],"APPROVE,":[116],"REJECT_REDUNDANT,":[117],"REJECT_CONFLICT":[119],"verdicts":[120],"for":[121,208,232,249],"each":[122,218],"candidate":[123,155],"entry.":[124],"Rejected-but-salvageable":[125],"undergo":[127],"reflective":[129],"abstraction":[130],"loop":[131],"progressively":[133],"elevates":[134],"level":[136],"generalization.":[138],"further":[140],"sandbox":[142],"consistency":[143],"verification,":[144],"mentally":[148],"replays":[149],"historical":[150],"scenarios":[152],"injected":[157],"detect":[159],"potential":[160],"failures":[162],"before":[163],"commitment.":[164],"Evaluated":[165],"ALFWorld":[167],"(continuous":[168],"5-round":[169],"reset,":[170],"670":[171],"scenarios),":[172],"WebShop":[173],"(200-task":[174],"pipeline),":[175],"and":[176],"MemoryAgentBench":[177],"FactConsolidation,":[178],"FGHM":[179],"reduces":[180],"long-term":[182],"pollution":[184],"rate":[185,197],"from$\\mathbf{3":[186],"8":[187],"\\%}$(naive":[188],"LLM":[189],"summarization)":[190],"to$\\mathbf{9":[191],"\\%}$while":[192],"maintaining":[193],"success":[196],"degradation":[198],"only":[200],"4":[201],"percentage":[202],"points":[203],"5":[205],"rounds-compared":[206],"to$20+$points":[207],"baselines.":[209],"Comprehensive":[210],"ablation":[211],"studies":[212],"confirm":[213],"independent":[215],"contribution":[216],"component.":[219],"additionally":[221],"propose":[222],"LTM":[223],"Pollution":[224],"Rate":[225],"(LPR)":[226],"new":[229],"evaluation":[230],"metric":[231],"quality":[234],"assessment,":[235],"complementing":[236],"existing":[237],"task-level":[238],"metrics.":[239],"Our":[240],"results":[241],"demonstrate":[242],"failure-aware":[244],"write":[245],"gating":[246],"essential":[248],"sustainable,":[250],"deployment.":[253]},"cited_by_api_url":"https://api.openalex.org/works?filter=cites:W7170034571","counts_by_year":[],"updated_date":"2026-11-20T13:11:24.684918","created_date":"2026-07-23T00:00:00"}
