{"id":"https://openalex.org/W4416078076","doi":"https://doi.org/10.1109/tai.2025.3630623","title":"ArMA: Mitigating Catastrophic Forgetting using Attention-Regularized Model Averaging in Continual Fine-tuning Large Language Models","display_name":"ArMA: Mitigating Catastrophic Forgetting using Attention-Regularized Model Averaging in Continual Fine-tuning Large Language Models","publication_year":2025,"publication_date":"2025-11-10","ids":{"openalex":"https://openalex.org/W4416078076","doi":"https://doi.org/10.1109/tai.2025.3630623"},"language":null,"primary_location":{"id":"doi:10.1109/tai.2025.3630623","is_oa":false,"landing_page_url":"https://doi.org/10.1109/tai.2025.3630623","pdf_url":null,"source":{"id":"https://openalex.org/S4210169448","display_name":"IEEE Transactions on Artificial Intelligence","issn_l":"2691-4581","issn":["2691-4581"],"is_oa":false,"is_in_doaj":false,"is_core":true,"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":"journal"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"IEEE Transactions on Artificial Intelligence","raw_type":"journal-article"},"type":"article","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/A5007950680","display_name":"Xihe Qiu","orcid":"https://orcid.org/0000-0003-4024-925X"},"institutions":[{"id":"https://openalex.org/I141962983","display_name":"Shanghai University of Engineering Science","ror":"https://ror.org/0557b9y08","country_code":"CN","type":"education","lineage":["https://openalex.org/I141962983"]}],"countries":["CN"],"is_corresponding":true,"raw_author_name":"Xihe Qiu","raw_affiliation_strings":["School of Electronic and Electrical Engineering, Shanghai University of Engineering Science, Shanghai, China"],"raw_orcid":"https://orcid.org/0000-0003-4024-925X","affiliations":[{"raw_affiliation_string":"School of Electronic and Electrical Engineering, Shanghai University of Engineering Science, Shanghai, China","institution_ids":["https://openalex.org/I141962983"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5103325266","display_name":"Leijun Cheng","orcid":"https://orcid.org/0009-0003-6625-1961"},"institutions":[{"id":"https://openalex.org/I141962983","display_name":"Shanghai University of Engineering Science","ror":"https://ror.org/0557b9y08","country_code":"CN","type":"education","lineage":["https://openalex.org/I141962983"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Leijun Cheng","raw_affiliation_strings":["School of Electronic and Electrical Engineering, Shanghai University of Engineering Science, Shanghai, China"],"raw_orcid":"https://orcid.org/0009-0003-6625-1961","affiliations":[{"raw_affiliation_string":"School of Electronic and Electrical Engineering, Shanghai University of Engineering Science, Shanghai, China","institution_ids":["https://openalex.org/I141962983"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5090685223","display_name":"Teqi Hao","orcid":"https://orcid.org/0009-0006-9510-8154"},"institutions":[{"id":"https://openalex.org/I141962983","display_name":"Shanghai University of Engineering Science","ror":"https://ror.org/0557b9y08","country_code":"CN","type":"education","lineage":["https://openalex.org/I141962983"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Teqi Hao","raw_affiliation_strings":["School of Electronic and Electrical Engineering, Shanghai University of Engineering Science, Shanghai, China"],"raw_orcid":"https://orcid.org/0009-0006-9510-8154","affiliations":[{"raw_affiliation_string":"School of Electronic and Electrical Engineering, Shanghai University of Engineering Science, Shanghai, China","institution_ids":["https://openalex.org/I141962983"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5102869238","display_name":"Xiaoyu Tan","orcid":"https://orcid.org/0000-0003-3555-7143"},"institutions":[{"id":"https://openalex.org/I2250653659","display_name":"Tencent (China)","ror":"https://ror.org/00hhjss72","country_code":"CN","type":"company","lineage":["https://openalex.org/I2250653659"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Xiaoyu Tan","raw_affiliation_strings":["Tencent Youtu Lab, Shanghai, China"],"raw_orcid":"https://orcid.org/0000-0003-3555-7143","affiliations":[{"raw_affiliation_string":"Tencent Youtu Lab, Shanghai, China","institution_ids":["https://openalex.org/I2250653659"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":4,"corresponding_author_ids":["https://openalex.org/A5007950680"],"corresponding_institution_ids":["https://openalex.org/I141962983"],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":{"value":0.17026228,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":"7","issue":"5","first_page":"2920","last_page":"2930"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.43799999356269836,"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/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.43799999356269836,"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":0.16979999840259552,"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"}},{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.1298000067472458,"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/forgetting","display_name":"Forgetting","score":0.7225000262260437},{"id":"https://openalex.org/keywords/heuristic","display_name":"Heuristic","score":0.6044999957084656},{"id":"https://openalex.org/keywords/task","display_name":"Task (project management)","score":0.552299976348877},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.475600004196167},{"id":"https://openalex.org/keywords/regularization","display_name":"Regularization (linguistics)","score":0.45899999141693115},{"id":"https://openalex.org/keywords/generalization","display_name":"Generalization","score":0.43880000710487366}],"concepts":[{"id":"https://openalex.org/C7149132","wikidata":"https://www.wikidata.org/wiki/Q1377840","display_name":"Forgetting","level":2,"score":0.7225000262260437},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6739000082015991},{"id":"https://openalex.org/C173801870","wikidata":"https://www.wikidata.org/wiki/Q201413","display_name":"Heuristic","level":2,"score":0.6044999957084656},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.552299976348877},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.475600004196167},{"id":"https://openalex.org/C2776135515","wikidata":"https://www.wikidata.org/wiki/Q17143721","display_name":"Regularization (linguistics)","level":2,"score":0.45899999141693115},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.4499000012874603},{"id":"https://openalex.org/C177148314","wikidata":"https://www.wikidata.org/wiki/Q170084","display_name":"Generalization","level":2,"score":0.43880000710487366},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4311000108718872},{"id":"https://openalex.org/C36503486","wikidata":"https://www.wikidata.org/wiki/Q11235244","display_name":"Domain (mathematical analysis)","level":2,"score":0.3781999945640564},{"id":"https://openalex.org/C112930515","wikidata":"https://www.wikidata.org/wiki/Q4389547","display_name":"Risk analysis (engineering)","level":1,"score":0.3612000048160553},{"id":"https://openalex.org/C2780009758","wikidata":"https://www.wikidata.org/wiki/Q6804172","display_name":"Measure (data warehouse)","level":2,"score":0.3237999975681305},{"id":"https://openalex.org/C177606310","wikidata":"https://www.wikidata.org/wiki/Q5674297","display_name":"Adaptability","level":2,"score":0.2874999940395355}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1109/tai.2025.3630623","is_oa":false,"landing_page_url":"https://doi.org/10.1109/tai.2025.3630623","pdf_url":null,"source":{"id":"https://openalex.org/S4210169448","display_name":"IEEE Transactions on Artificial Intelligence","issn_l":"2691-4581","issn":["2691-4581"],"is_oa":false,"is_in_doaj":false,"is_core":true,"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":"journal"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"IEEE Transactions on Artificial Intelligence","raw_type":"journal-article"}],"best_oa_location":null,"sustainable_development_goals":[],"awards":[{"id":"https://openalex.org/G3593522634","display_name":null,"funder_award_id":"23ZR1425400","funder_id":"https://openalex.org/F4320309612","funder_display_name":"Natural Science Foundation of Shanghai"}],"funders":[{"id":"https://openalex.org/F4320309612","display_name":"Natural Science Foundation of Shanghai","ror":null}],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":29,"referenced_works":["https://openalex.org/W2100805904","https://openalex.org/W2319803923","https://openalex.org/W2546786066","https://openalex.org/W2560647685","https://openalex.org/W2743151379","https://openalex.org/W2945383715","https://openalex.org/W3105882417","https://openalex.org/W3178522238","https://openalex.org/W3212604410","https://openalex.org/W4313492042","https://openalex.org/W4384211302","https://openalex.org/W4385570738","https://openalex.org/W4385571213","https://openalex.org/W4385572119","https://openalex.org/W4385573164","https://openalex.org/W4386076061","https://openalex.org/W4389519598","https://openalex.org/W4389519807","https://openalex.org/W4389520255","https://openalex.org/W4389523949","https://openalex.org/W4390035079","https://openalex.org/W4400933678","https://openalex.org/W4401057236","https://openalex.org/W4402353887","https://openalex.org/W4404574611","https://openalex.org/W4404782827","https://openalex.org/W4404783028","https://openalex.org/W4407601528","https://openalex.org/W7131834546"],"related_works":[],"abstract_inverted_index":{"Recent":[0],"advancements":[1],"in":[2,33,98,181],"continual":[3,100,183],"fine-tuning":[4,68,101,184],"have":[5],"aimed":[6],"to":[7,48,75,77,117,135],"enhance":[8],"the":[9,64,73,99,109,121,147,156,175,182,190],"instruction-following":[10],"capabilities":[11,180],"of":[12,67,102,185],"large":[13],"language":[14],"models":[15,22,150],"(LLMs)":[16],"within":[17],"domain-specific":[18,200],"contexts.":[19],"However,":[20],"these":[21],"often":[23],"suffer":[24],"from":[25],"catastrophic":[26,96],"forgetting,":[27],"manifesting":[28],"as":[29],"a":[30,41,52,89,129],"significant":[31],"decline":[32],"performance":[34,50],"on":[35,51,69,84,132,146],"general":[36,122,139,191],"domain":[37,54,192],"tasks.":[38,61,141],"This":[39,93,142],"presents":[40],"substantial":[42],"challenge":[43],"for":[44],"developers":[45],"who":[46],"seek":[47],"improve":[49],"specific":[53],"without":[55],"compromising":[56],"efficacy":[57],"across":[58,166],"previously":[59],"established":[60],"To":[62],"mitigate":[63],"negative":[65],"impact":[66],"generalization":[70],"and":[71,119,187],"enable":[72],"model":[74,106,111,172],"adapt":[76],"new":[78],"tasks":[79],"while":[80],"preserving":[81],"its":[82],"ability":[83],"general-domain":[85],"tasks,":[86],"we":[87],"propose":[88],"novel":[90],"framework,":[91],"ARMA.":[92],"framework":[94,127],"addresses":[95],"forgetting":[97],"LLMs":[103,186],"through":[104],"attention-regularized":[105,171],"averaging.":[107],"Unlike":[108],"typical":[110],"average,":[112],"which":[113,194],"utilizes":[114],"various":[115],"metrics":[116],"measure":[118],"balance":[120],"capabilities,":[123,193],"our":[124],"proposed":[125],"ARMA":[126],"employs":[128],"heuristic":[130],"based":[131],"attention":[133,153],"regularization":[134],"universally":[136],"consider":[137],"all":[138],"instruction":[140],"approach":[143],"is":[144],"predicated":[145],"hypothesis":[148],"that":[149,170],"exhibiting":[151],"similar":[152],"distributions":[154],"over":[155],"input":[157],"instructions":[158],"will":[159,195],"likely":[160],"yield":[161],"comparable":[162],"outcomes.":[163],"Our":[164],"evaluation":[165],"multiple":[167],"benchmarks":[168],"demonstrates":[169],"averaging":[173],"balances":[174],"trade-offs":[176],"between":[177],"different":[178],"task":[179],"effectively":[188],"reserves":[189],"potentially":[196],"benefit":[197],"complex":[198],"industrial":[199],"applications.":[201]},"counts_by_year":[],"updated_date":"2026-05-05T06:06:40.768181","created_date":"2025-11-10T00:00:00"}