{"id":"https://openalex.org/W4416034765","doi":"https://doi.org/10.18653/v1/2025.findings-emnlp.578","title":"Mitigating Attention Localization in Small Scale: Self-Attention Refinement via One-step Belief Propagation","display_name":"Mitigating Attention Localization in Small Scale: Self-Attention Refinement via One-step Belief Propagation","publication_year":2025,"publication_date":"2025-01-01","ids":{"openalex":"https://openalex.org/W4416034765","doi":"https://doi.org/10.18653/v1/2025.findings-emnlp.578"},"language":null,"primary_location":{"id":"doi:10.18653/v1/2025.findings-emnlp.578","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-emnlp.578","pdf_url":"https://aclanthology.org/2025.findings-emnlp.578.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: EMNLP 2025","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2025.findings-emnlp.578.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5023777659","display_name":"Nakyung Lee","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Nakyung Lee","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Yeongoon Kim","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yeongoon Kim","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100605347","display_name":"M. Matthew Oh","orcid":"https://orcid.org/0000-0002-6702-5785"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Minhae Oh","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5080332152","display_name":"Suhwan Kim","orcid":"https://orcid.org/0000-0001-9107-2963"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Suhwan Kim","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5104546494","display_name":"Ja Ho Koo","orcid":"https://orcid.org/0000-0003-3482-625X"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jin Woo Koo","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5087097952","display_name":"Hye-Won Jo","orcid":"https://orcid.org/0000-0002-0114-4212"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Hyewon Jo","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5100376250","display_name":"Jungwoo Lee","orcid":"https://orcid.org/0000-0001-6689-4500"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jungwoo Lee","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":true,"cited_by_count":0,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"10897","last_page":"10912"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11605","display_name":"Visual Attention and Saliency Detection","score":0.13979999721050262,"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"}},"topics":[{"id":"https://openalex.org/T11605","display_name":"Visual Attention and Saliency Detection","score":0.13979999721050262,"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/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.13359999656677246,"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/T11094","display_name":"Face Recognition and Perception","score":0.08990000188350677,"subfield":{"id":"https://openalex.org/subfields/2805","display_name":"Cognitive Neuroscience"},"field":{"id":"https://openalex.org/fields/28","display_name":"Neuroscience"},"domain":{"id":"https://openalex.org/domains/1","display_name":"Life Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/noise","display_name":"Noise (video)","score":0.26669999957084656},{"id":"https://openalex.org/keywords/sequence","display_name":"Sequence (biology)","score":0.25429999828338623},{"id":"https://openalex.org/keywords/feature","display_name":"Feature (linguistics)","score":0.2535000145435333},{"id":"https://openalex.org/keywords/probabilistic-logic","display_name":"Probabilistic logic","score":0.25099998712539673},{"id":"https://openalex.org/keywords/minification","display_name":"Minification","score":0.2451999932527542}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5414999723434448},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.43799999356269836},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.31709998846054077},{"id":"https://openalex.org/C99498987","wikidata":"https://www.wikidata.org/wiki/Q2210247","display_name":"Noise (video)","level":3,"score":0.26669999957084656},{"id":"https://openalex.org/C2778112365","wikidata":"https://www.wikidata.org/wiki/Q3511065","display_name":"Sequence (biology)","level":2,"score":0.25429999828338623},{"id":"https://openalex.org/C2776401178","wikidata":"https://www.wikidata.org/wiki/Q12050496","display_name":"Feature (linguistics)","level":2,"score":0.2535000145435333},{"id":"https://openalex.org/C49937458","wikidata":"https://www.wikidata.org/wiki/Q2599292","display_name":"Probabilistic logic","level":2,"score":0.25099998712539673},{"id":"https://openalex.org/C147764199","wikidata":"https://www.wikidata.org/wiki/Q6865248","display_name":"Minification","level":2,"score":0.2451999932527542},{"id":"https://openalex.org/C152948882","wikidata":"https://www.wikidata.org/wiki/Q4060686","display_name":"Belief propagation","level":3,"score":0.2443999946117401},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.24250000715255737}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2025.findings-emnlp.578","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-emnlp.578","pdf_url":"https://aclanthology.org/2025.findings-emnlp.578.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: EMNLP 2025","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2025.findings-emnlp.578","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-emnlp.578","pdf_url":"https://aclanthology.org/2025.findings-emnlp.578.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: EMNLP 2025","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[{"id":"https://openalex.org/G1451011034","display_name":null,"funder_award_id":"2021-II212068","funder_id":"https://openalex.org/F4320328359","funder_display_name":"Ministry of Science and ICT, South Korea"},{"id":"https://openalex.org/G30685149","display_name":null,"funder_award_id":"BK21 FOUR","funder_id":"https://openalex.org/F4320320671","funder_display_name":"National Research Foundation"},{"id":"https://openalex.org/G3598079552","display_name":null,"funder_award_id":"RS-2021-II211343","funder_id":"https://openalex.org/F4320335489","funder_display_name":"Institute for Information and Communications Technology Promotion"},{"id":"https://openalex.org/G45824782","display_name":null,"funder_award_id":"RS-2024-00413957","funder_id":"https://openalex.org/F4320320671","funder_display_name":"National Research Foundation"},{"id":"https://openalex.org/G5562307789","display_name":null,"funder_award_id":"BK21 FOUR","funder_id":"https://openalex.org/F4320322120","funder_display_name":"National Research Foundation of Korea"},{"id":"https://openalex.org/G6390571125","display_name":null,"funder_award_id":"RS-2024-00413957","funder_id":"https://openalex.org/F4320328359","funder_display_name":"Ministry of Science and ICT, South Korea"},{"id":"https://openalex.org/G7073589060","display_name":null,"funder_award_id":"RS-2024-00413957","funder_id":"https://openalex.org/F4320322120","funder_display_name":"National Research Foundation of Korea"},{"id":"https://openalex.org/G7288789209","display_name":null,"funder_award_id":"RS-2021-II211343","funder_id":"https://openalex.org/F4320328359","funder_display_name":"Ministry of Science and ICT, South Korea"},{"id":"https://openalex.org/G7575369263","display_name":null,"funder_award_id":"RS-2021-II212068","funder_id":"https://openalex.org/F4320335489","funder_display_name":"Institute for Information and Communications Technology Promotion"},{"id":"https://openalex.org/G7699972493","display_name":null,"funder_award_id":"RS-2021-II212068","funder_id":"https://openalex.org/F4320328359","funder_display_name":"Ministry of Science and ICT, South Korea"},{"id":"https://openalex.org/G7798134731","display_name":null,"funder_award_id":"RS-2021-II211343","funder_id":"https://openalex.org/F4320321292","funder_display_name":"Seoul National University"}],"funders":[{"id":"https://openalex.org/F4320320671","display_name":"National Research Foundation","ror":"https://ror.org/05s0g1g46"},{"id":"https://openalex.org/F4320321292","display_name":"Seoul National University","ror":"https://ror.org/04h9pn542"},{"id":"https://openalex.org/F4320322120","display_name":"National Research Foundation of Korea","ror":"https://ror.org/013aysd81"},{"id":"https://openalex.org/F4320328359","display_name":"Ministry of Science and ICT, South Korea","ror":"https://ror.org/01wpjm123"},{"id":"https://openalex.org/F4320335489","display_name":"Institute for Information and Communications Technology Promotion","ror":"https://ror.org/01g0hqq23"}],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4416034765.pdf","grobid_xml":"https://content.openalex.org/works/W4416034765.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Transformer-based":[0],"self-attention":[1],"mechanism":[2],"serves":[3],"as":[4],"the":[5,69,76],"core":[6],"of":[7,24,72],"modern":[8],"language":[9],"models,":[10,109],"yet":[11],"it":[12],"often":[13],"suffers":[14],"from":[15],"localization,":[16],"where":[17],"attentions":[18],"collapse":[19,86],"onto":[20],"a":[21,42,50],"limited":[22],"subset":[23],"tokens":[25],"and":[26,57,90],"fail":[27],"to":[28],"capture":[29],"long-range":[30],"dependencies.To":[31],"address":[32],"this":[33],"issue,":[34],"we":[35,61,103],"propose":[36],"Self-Attention":[37],"One-step":[38],"Belief":[39],"Propagation":[40],"(SAOBP),":[41],"refinement":[43],"framework":[44],"that":[45,67,81],"injects":[46],"multi-hop":[47],"relationships":[48],"through":[49],"belief":[51],"propagation":[52],"process":[53],"1":[54],".To":[55],"interpret":[56],"quantify":[58],"these":[59],"interactions,":[60],"introduce":[62],"Global":[63],"Token":[64],"Dependency":[65],"(GTD)":[66],"captures":[68],"relative":[70],"contribution":[71],"multihop":[73],"connections":[74],"within":[75],"attention":[77],"graph.Empirical":[78],"results":[79],"indicate":[80],"SAOBP":[82],"helps":[83],"prevent":[84],"entropy":[85],"in":[87,100,107,117],"deeper":[88],"layers":[89],"adaptively":[91],"maintains":[92],"GTD":[93],"at":[94],"task-appropriate":[95],"levels,":[96],"thereby":[97],"supporting":[98],"improvements":[99],"model":[101],"performance.Importantly,":[102],"observe":[104],"competitive":[105],"gains":[106],"small-scale":[108],"highlighting":[110],"its":[111],"potential":[112],"for":[113],"improving":[114],"inference":[115],"quality":[116],"resource-constrained":[118],"scenarios.":[119]},"counts_by_year":[],"updated_date":"2026-07-23T08:03:31.855105","created_date":"2025-11-08T00:00:00"}