{"id":"https://openalex.org/W4416036311","doi":"https://doi.org/10.18653/v1/2025.emnlp-main.742","title":"Calibration Across Layers: Understanding Calibration Evolution in LLMs","display_name":"Calibration Across Layers: Understanding Calibration Evolution in LLMs","publication_year":2025,"publication_date":"2025-01-01","ids":{"openalex":"https://openalex.org/W4416036311","doi":"https://doi.org/10.18653/v1/2025.emnlp-main.742"},"language":null,"primary_location":{"id":"doi:10.18653/v1/2025.emnlp-main.742","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.emnlp-main.742","pdf_url":"https://aclanthology.org/2025.emnlp-main.742.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":"Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing","raw_type":"proceedings-article"},"type":"article","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2025.emnlp-main.742.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5103253154","display_name":"Abhinav Joshi","orcid":"https://orcid.org/0000-0001-6756-1126"},"institutions":[{"id":"https://openalex.org/I94234084","display_name":"Indian Institute of Technology Kanpur","ror":"https://ror.org/05pjsgx75","country_code":"IN","type":"education","lineage":["https://openalex.org/I94234084"]}],"countries":["IN"],"is_corresponding":true,"raw_author_name":"Abhinav Joshi","raw_affiliation_strings":["Department of Computer Science and Engineering Indian Institute of Technology Kanpur (IIT Kanpur ) Microsoft Research Blog "],"affiliations":[{"raw_affiliation_string":"Department of Computer Science and Engineering Indian Institute of Technology Kanpur (IIT Kanpur ) Microsoft Research Blog ","institution_ids":["https://openalex.org/I94234084"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5102533871","display_name":"Areeb Ahmad","orcid":null},"institutions":[{"id":"https://openalex.org/I94234084","display_name":"Indian Institute of Technology Kanpur","ror":"https://ror.org/05pjsgx75","country_code":"IN","type":"education","lineage":["https://openalex.org/I94234084"]}],"countries":["IN"],"is_corresponding":false,"raw_author_name":"Areeb Ahmad","raw_affiliation_strings":["Department of Computer Science and Engineering Indian Institute of Technology Kanpur (IIT Kanpur ) Microsoft Research Blog "],"affiliations":[{"raw_affiliation_string":"Department of Computer Science and Engineering Indian Institute of Technology Kanpur (IIT Kanpur ) Microsoft Research Blog ","institution_ids":["https://openalex.org/I94234084"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5076043215","display_name":"Ashutosh Modi","orcid":"https://orcid.org/0000-0002-0962-8350"},"institutions":[{"id":"https://openalex.org/I94234084","display_name":"Indian Institute of Technology Kanpur","ror":"https://ror.org/05pjsgx75","country_code":"IN","type":"education","lineage":["https://openalex.org/I94234084"]}],"countries":["IN"],"is_corresponding":false,"raw_author_name":"Ashutosh Modi","raw_affiliation_strings":["Department of Computer Science and Engineering Indian Institute of Technology Kanpur (IIT Kanpur ) Microsoft Research Blog "],"affiliations":[{"raw_affiliation_string":"Department of Computer Science and Engineering Indian Institute of Technology Kanpur (IIT Kanpur ) Microsoft Research Blog ","institution_ids":["https://openalex.org/I94234084"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":3,"corresponding_author_ids":["https://openalex.org/A5103253154"],"corresponding_institution_ids":["https://openalex.org/I94234084"],"apc_list":null,"apc_paid":null,"fwci":0.7034,"has_fulltext":true,"cited_by_count":1,"citation_normalized_percentile":{"value":0.76110759,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":{"min":97,"max":99},"biblio":{"volume":null,"issue":null,"first_page":"14697","last_page":"14725"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10412","display_name":"Microfluidic and Capillary Electrophoresis Applications","score":0.05959999933838844,"subfield":{"id":"https://openalex.org/subfields/2204","display_name":"Biomedical Engineering"},"field":{"id":"https://openalex.org/fields/22","display_name":"Engineering"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T10412","display_name":"Microfluidic and Capillary Electrophoresis Applications","score":0.05959999933838844,"subfield":{"id":"https://openalex.org/subfields/2204","display_name":"Biomedical Engineering"},"field":{"id":"https://openalex.org/fields/22","display_name":"Engineering"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T11367","display_name":"Particle accelerators and beam dynamics","score":0.01899999938905239,"subfield":{"id":"https://openalex.org/subfields/2202","display_name":"Aerospace Engineering"},"field":{"id":"https://openalex.org/fields/22","display_name":"Engineering"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T12300","display_name":"Advanced Electrical Measurement Techniques","score":0.016499999910593033,"subfield":{"id":"https://openalex.org/subfields/2208","display_name":"Electrical and Electronic Engineering"},"field":{"id":"https://openalex.org/fields/22","display_name":"Engineering"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/calibration","display_name":"Calibration","score":0.7142999768257141},{"id":"https://openalex.org/keywords/tracking","display_name":"Tracking (education)","score":0.23409999907016754},{"id":"https://openalex.org/keywords/noise","display_name":"Noise (video)","score":0.22380000352859497}],"concepts":[{"id":"https://openalex.org/C165838908","wikidata":"https://www.wikidata.org/wiki/Q736777","display_name":"Calibration","level":2,"score":0.7142999768257141},{"id":"https://openalex.org/C62649853","wikidata":"https://www.wikidata.org/wiki/Q199687","display_name":"Remote sensing","level":1,"score":0.5371999740600586},{"id":"https://openalex.org/C39432304","wikidata":"https://www.wikidata.org/wiki/Q188847","display_name":"Environmental science","level":0,"score":0.5254999995231628},{"id":"https://openalex.org/C49204034","wikidata":"https://www.wikidata.org/wiki/Q52139","display_name":"Climatology","level":1,"score":0.3086000084877014},{"id":"https://openalex.org/C153294291","wikidata":"https://www.wikidata.org/wiki/Q25261","display_name":"Meteorology","level":1,"score":0.28450000286102295},{"id":"https://openalex.org/C205649164","wikidata":"https://www.wikidata.org/wiki/Q1071","display_name":"Geography","level":0,"score":0.26669999957084656},{"id":"https://openalex.org/C127313418","wikidata":"https://www.wikidata.org/wiki/Q1069","display_name":"Geology","level":0,"score":0.24320000410079956},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.23579999804496765},{"id":"https://openalex.org/C2775936607","wikidata":"https://www.wikidata.org/wiki/Q466845","display_name":"Tracking (education)","level":2,"score":0.23409999907016754},{"id":"https://openalex.org/C99498987","wikidata":"https://www.wikidata.org/wiki/Q2210247","display_name":"Noise (video)","level":3,"score":0.22380000352859497}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2025.emnlp-main.742","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.emnlp-main.742","pdf_url":"https://aclanthology.org/2025.emnlp-main.742.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":"Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2025.emnlp-main.742","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.emnlp-main.742","pdf_url":"https://aclanthology.org/2025.emnlp-main.742.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":"Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[],"funders":[{"id":"https://openalex.org/F4320320712","display_name":"Indian Institute of Technology Kanpur","ror":"https://ror.org/05pjsgx75"}],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4416036311.pdf","grobid_xml":"https://content.openalex.org/works/W4416036311.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Large":[0],"Language":[1],"Models":[2],"(LLMs)":[3],"have":[4,27],"demonstrated":[5],"inherent":[6],"calibration":[7,58,98,108,119],"capabilities,":[8],"where":[9,82],"predicted":[10],"probabilities":[11],"align":[12],"well":[13],"with":[14],"correctness,":[15],"despite":[16],"prior":[17],"findings":[18,116],"that":[19,118],"deep":[20],"neural":[21],"networks":[22],"are":[23],"often":[24],"overconfident.Recent":[25],"studies":[26],"linked":[28],"this":[29,48],"behavior":[30],"to":[31],"specific":[32],"components":[33],"in":[34,78,100,132],"the":[35,43,61,68,79,101,126],"final":[36,134],"layer,":[37],"such":[38],"as":[39],"entropy":[40],"neurons":[41],"and":[42,111],"unembedding":[44],"matrix's":[45],"null":[46],"space.In":[47],"work,":[49],"we":[50,71,94],"provide":[51],"a":[52,73,96,121],"complementary":[53],"perspective":[54],"by":[55],"investigating":[56],"how":[57,140],"evolves":[59],"throughout":[60,125],"network's":[62,127],"depth.Analyzing":[63],"multiple":[64],"open-weight":[65],"models":[66],"on":[67],"MMLU":[69],"benchmark,":[70],"uncover":[72],"distinct":[74],"confidence":[75,84],"correction":[76],"phase":[77],"upper/later":[80],"layers,":[81],"model":[83],"is":[85,120],"actively":[86],"recalibrated":[87],"after":[88],"decision":[89],"certainty":[90],"has":[91],"been":[92],"reached.Furthermore,":[93],"identify":[95],"low-dimensional":[97],"direction":[99],"residual":[102],"stream":[103],"whose":[104],"perturbation":[105],"significantly":[106],"improves":[107],"metrics":[109],"(ECE":[110],"MCE)":[112],"without":[113],"harming":[114],"accuracy.Our":[115],"suggest":[117],"distributed":[122],"phenomenon,":[123],"shaped":[124],"forward":[128],"pass,":[129],"not":[130],"just":[131],"its":[133],"projection,":[135],"providing":[136],"new":[137],"insights":[138],"into":[139],"confidence-regulating":[141],"mechanisms":[142],"operate":[143],"within":[144],"LLMs.":[145]},"counts_by_year":[{"year":2026,"cited_by_count":1}],"updated_date":"2026-03-20T23:20:44.827607","created_date":"2025-11-08T00:00:00"}