{"id":"https://openalex.org/W7166889768","doi":"https://doi.org/10.18653/v1/2026.findings-acl.133","title":"Where CoT Reasoning Commits: Entropy Traces Identify Interpretable Attention Heads","display_name":"Where CoT Reasoning Commits: Entropy Traces Identify Interpretable Attention Heads","publication_year":2026,"publication_date":"2026-01-01","ids":{"openalex":"https://openalex.org/W7166889768","doi":"https://doi.org/10.18653/v1/2026.findings-acl.133"},"language":null,"primary_location":{"id":"doi:10.18653/v1/2026.findings-acl.133","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-acl.133","pdf_url":"https://aclanthology.org/2026.findings-acl.133.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: ACL 2026","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2026.findings-acl.133.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5138708172","display_name":"Tianhe Zhang","orcid":null},"institutions":[{"id":"https://openalex.org/I125839683","display_name":"Beijing Institute of Technology","ror":"https://ror.org/01skt4w74","country_code":"CN","type":"education","lineage":["https://openalex.org/I125839683","https://openalex.org/I890469752"]},{"id":"https://openalex.org/I202334528","display_name":"Beijing Electronic Science and Technology Institute","ror":"https://ror.org/01xdzh226","country_code":"CN","type":"education","lineage":["https://openalex.org/I202334528"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Tianhe Zhang","raw_affiliation_strings":["School of Computer Science and Technology , Beijing Institute of Technology , Beijing , China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"School of Computer Science and Technology , Beijing Institute of Technology , Beijing , China","institution_ids":["https://openalex.org/I125839683","https://openalex.org/I202334528"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139774999","display_name":"Yonghong Deng","orcid":null},"institutions":[{"id":"https://openalex.org/I125839683","display_name":"Beijing Institute of Technology","ror":"https://ror.org/01skt4w74","country_code":"CN","type":"education","lineage":["https://openalex.org/I125839683","https://openalex.org/I890469752"]},{"id":"https://openalex.org/I202334528","display_name":"Beijing Electronic Science and Technology Institute","ror":"https://ror.org/01xdzh226","country_code":"CN","type":"education","lineage":["https://openalex.org/I202334528"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Yonghong Deng","raw_affiliation_strings":["School of Computer Science and Technology , Beijing Institute of Technology , Beijing , China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"School of Computer Science and Technology , Beijing Institute of Technology , Beijing , China","institution_ids":["https://openalex.org/I125839683","https://openalex.org/I202334528"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139787893","display_name":"Ping Jian","orcid":null},"institutions":[{"id":"https://openalex.org/I125839683","display_name":"Beijing Institute of Technology","ror":"https://ror.org/01skt4w74","country_code":"CN","type":"education","lineage":["https://openalex.org/I125839683","https://openalex.org/I890469752"]},{"id":"https://openalex.org/I202334528","display_name":"Beijing Electronic Science and Technology Institute","ror":"https://ror.org/01xdzh226","country_code":"CN","type":"education","lineage":["https://openalex.org/I202334528"]},{"id":"https://openalex.org/I4210139765","display_name":"Beijing Computing Center","ror":"https://ror.org/047r47y76","country_code":"CN","type":"other","lineage":["https://openalex.org/I4210139765"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Ping Jian","raw_affiliation_strings":["Beijing Engineering Research Center of High Volume Language Information Processing and Cloud Computing Applications , Beijing Institute of Technology , Beijing , China","School of Computer Science and Technology , Beijing Institute of Technology , Beijing , China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Beijing Engineering Research Center of High Volume Language Information Processing and Cloud Computing Applications , Beijing Institute of Technology , Beijing , China","institution_ids":["https://openalex.org/I4210139765"]},{"raw_affiliation_string":"School of Computer Science and Technology , Beijing Institute of Technology , Beijing , China","institution_ids":["https://openalex.org/I125839683","https://openalex.org/I202334528"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139732123","display_name":"Zhen Yang","orcid":null},"institutions":[{"id":"https://openalex.org/I125839683","display_name":"Beijing Institute of Technology","ror":"https://ror.org/01skt4w74","country_code":"CN","type":"education","lineage":["https://openalex.org/I125839683","https://openalex.org/I890469752"]},{"id":"https://openalex.org/I202334528","display_name":"Beijing Electronic Science and Technology Institute","ror":"https://ror.org/01xdzh226","country_code":"CN","type":"education","lineage":["https://openalex.org/I202334528"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Zhen Yang","raw_affiliation_strings":["School of Computer Science and Technology , Beijing Institute of Technology , Beijing , China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"School of Computer Science and Technology , Beijing Institute of Technology , Beijing , China","institution_ids":["https://openalex.org/I125839683","https://openalex.org/I202334528"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139757447","display_name":"Boyang Wang","orcid":null},"institutions":[{"id":"https://openalex.org/I125839683","display_name":"Beijing Institute of Technology","ror":"https://ror.org/01skt4w74","country_code":"CN","type":"education","lineage":["https://openalex.org/I125839683","https://openalex.org/I890469752"]},{"id":"https://openalex.org/I202334528","display_name":"Beijing Electronic Science and Technology Institute","ror":"https://ror.org/01xdzh226","country_code":"CN","type":"education","lineage":["https://openalex.org/I202334528"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Boyang Wang","raw_affiliation_strings":["School of Computer Science and Technology , Beijing Institute of Technology , Beijing , China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"School of Computer Science and Technology , Beijing Institute of Technology , Beijing , China","institution_ids":["https://openalex.org/I125839683","https://openalex.org/I202334528"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5139758007","display_name":"Xinyue Zhang","orcid":null},"institutions":[{"id":"https://openalex.org/I125839683","display_name":"Beijing Institute of Technology","ror":"https://ror.org/01skt4w74","country_code":"CN","type":"education","lineage":["https://openalex.org/I125839683","https://openalex.org/I890469752"]},{"id":"https://openalex.org/I202334528","display_name":"Beijing Electronic Science and Technology Institute","ror":"https://ror.org/01xdzh226","country_code":"CN","type":"education","lineage":["https://openalex.org/I202334528"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Xinyue Zhang","raw_affiliation_strings":["School of Computer Science and Technology , Beijing Institute of Technology , Beijing , China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"School of Computer Science and Technology , Beijing Institute of Technology , Beijing , China","institution_ids":["https://openalex.org/I125839683","https://openalex.org/I202334528"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":3,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":{"value":0.84657689,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"2777","last_page":"2795"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":0.722000002861023,"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/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":0.722000002861023,"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":0.03099999949336052,"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":0.022299999371170998,"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/entropy","display_name":"Entropy (arrow of time)","score":0.5214999914169312},{"id":"https://openalex.org/keywords/pattern-recognition","display_name":"Pattern recognition (psychology)","score":0.36059999465942383},{"id":"https://openalex.org/keywords/principle-of-maximum-entropy","display_name":"Principle of maximum entropy","score":0.22509999573230743},{"id":"https://openalex.org/keywords/kullback\u2013leibler-divergence","display_name":"Kullback\u2013Leibler divergence","score":0.2214999943971634}],"concepts":[{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5669999718666077},{"id":"https://openalex.org/C106301342","wikidata":"https://www.wikidata.org/wiki/Q4117933","display_name":"Entropy (arrow of time)","level":2,"score":0.5214999914169312},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5181000232696533},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.36059999465942383},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.3292999863624573},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.2694000005722046},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.23510000109672546},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.22910000383853912},{"id":"https://openalex.org/C9679016","wikidata":"https://www.wikidata.org/wiki/Q1417473","display_name":"Principle of maximum entropy","level":2,"score":0.22509999573230743},{"id":"https://openalex.org/C171752962","wikidata":"https://www.wikidata.org/wiki/Q255166","display_name":"Kullback\u2013Leibler divergence","level":2,"score":0.2214999943971634}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2026.findings-acl.133","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-acl.133","pdf_url":"https://aclanthology.org/2026.findings-acl.133.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: ACL 2026","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2026.findings-acl.133","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-acl.133","pdf_url":"https://aclanthology.org/2026.findings-acl.133.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: ACL 2026","raw_type":"proceedings-article"},"sustainable_development_goals":[{"score":0.49110978841781616,"id":"https://metadata.un.org/sdg/10","display_name":"Reduced inequalities"}],"awards":[{"id":"https://openalex.org/G917634060","display_name":null,"funder_award_id":"62376130","funder_id":"https://openalex.org/F4320321001","funder_display_name":"National Natural Science Foundation of China"}],"funders":[{"id":"https://openalex.org/F4320321001","display_name":"National Natural Science Foundation of China","ror":"https://ror.org/01h0zpd94"}],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7166889768.pdf","grobid_xml":"https://content.openalex.org/works/W7166889768.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"While":[0],"LLMs":[1],"demonstrate":[2],"impressive":[3],"reasoning":[4],"capabilities,":[5],"their":[6],"internal":[7],"decision":[8],"dynamics":[9],"remain":[10],"opaque.To":[11],"render":[12],"these":[13,91,107],"process":[14],"interpretable":[15],"and":[16,42,60,72,80,100,117,140,191],"intervenable,":[17],"we":[18,93],"propose":[19],"Dynamic":[20],"Entropy":[21],"Tracing,":[22],"a":[23,62,68,88,95,111,121,162,169],"mechanism-aware":[24],"framework":[25],"that":[26,181,188],"interprets":[27],"the":[28,115,134,172,179,183],"evolving":[29],"\"choice":[30],"state\"":[31],"of":[32,97,133],"attention":[33,51],"heads":[34,109],"during":[35,194],"CoT":[36],"generation":[37],"through":[38],"stepwise":[39],"head-wise":[40],"option-logit":[41],"entropy":[43,59,79],"tracing.Our":[44],"analysis":[45],"reveals":[46],"distinct":[47],"functional":[48],"behaviors":[49],"at":[50],"heads:":[52],"Steadfast":[53,150],"Heads,":[54,74],"characterized":[55,75],"by":[56,76,137],"consistently":[57,77],"low":[58],"producing":[61,81],"sharp,":[63],"optionselective":[64],"logit":[65],"pattern":[66],"with":[67],"stable":[69],"top":[70],"choice,":[71],"Wavering":[73,128],"high":[78],"flat":[82],"or":[83],"oscillatory":[84],"option":[85],"logits":[86],"without":[87],"persistent":[89],"winner.Leveraging":[90],"traces,":[92],"identify":[94],"set":[96],"intervention":[98],"targets":[99],"perform":[101],"Selective":[102],"Head":[103],"Fine-Tuning,":[104],"updating":[105],"solely":[106],"selected":[108],"against":[110],"frozen":[112],"backbone.Experiments":[113],"across":[114],"LLaMA":[116],"Qwen":[118],"families":[119],"reveal":[120],"striking":[122],"plasticity":[123],"hierarchy:":[124],"fine-tuning":[125],"just":[126],"30":[127],"Heads":[129,151],"recovers":[130],"over":[131],"98%":[132],"performance":[135],"achieved":[136],"full-parameter":[138],"tuning,":[139],"in":[141],"some":[142],"settings":[143],"modestly":[144],"exceeds":[145],"it.In":[146],"contrast,":[147],"intervening":[148],"on":[149],"yields":[152],"much":[153],"less":[154],"gains.Our":[155],"findings":[156],"translate":[157],"processlevel":[158],"mechanistic":[159],"observables":[160],"into":[161],"principled":[163],"criterion":[164],"for":[165],"selective":[166],"fine-tuning,":[167],"offering":[168],"fundamental":[170],"insight:":[171],"most":[173],"effective":[174],"tuning":[175],"knobs":[176],"are":[177],"not":[178],"components":[180],"signal":[182],"final":[184],"decision,":[185],"but":[186],"those":[187],"retain":[189],"uncertainty,":[190],"thus":[192],"plasticity,":[193],"its":[195],"formation.1":[196]},"counts_by_year":[],"updated_date":"2026-08-26T07:47:46.906454","created_date":"2026-07-02T00:00:00"}