{"id":"https://openalex.org/W7137978284","doi":"https://doi.org/10.1609/aaai.v40i11.37913","title":"Distillation Dynamics: Towards Understanding Feature-Based Distillation in Vision Transformers","display_name":"Distillation Dynamics: Towards Understanding Feature-Based Distillation in Vision Transformers","publication_year":2026,"publication_date":"2026-03-14","ids":{"openalex":"https://openalex.org/W7137978284","doi":"https://doi.org/10.1609/aaai.v40i11.37913"},"language":null,"primary_location":{"id":"doi:10.1609/aaai.v40i11.37913","is_oa":true,"landing_page_url":"https://doi.org/10.1609/aaai.v40i11.37913","pdf_url":null,"source":{"id":"https://openalex.org/S4210191458","display_name":"Proceedings of the AAAI Conference on Artificial Intelligence","issn_l":"2159-5399","issn":["2159-5399","2374-3468"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/P4310320058","host_organization_name":"Association for the Advancement of Artificial Intelligence","host_organization_lineage":["https://openalex.org/P4310320058"],"host_organization_lineage_names":["Association for the Advancement of Artificial Intelligence"],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the AAAI Conference on Artificial Intelligence","raw_type":"journal-article"},"type":"article","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"diamond","oa_url":"https://doi.org/10.1609/aaai.v40i11.37913","any_repository_has_fulltext":false},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5129694094","display_name":"Huiyuan Tian","orcid":null},"institutions":[{"id":"https://openalex.org/I76130692","display_name":"Zhejiang University","ror":"https://ror.org/00a2xv884","country_code":"CN","type":"education","lineage":["https://openalex.org/I76130692"]}],"countries":["CN"],"is_corresponding":true,"raw_author_name":"Huiyuan Tian","raw_affiliation_strings":["Zhejiang University"],"affiliations":[{"raw_affiliation_string":"Zhejiang University","institution_ids":["https://openalex.org/I76130692"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5129728954","display_name":"Bonan Xu","orcid":null},"institutions":[{"id":"https://openalex.org/I14243506","display_name":"Hong Kong Polytechnic University","ror":"https://ror.org/0030zas98","country_code":"HK","type":"education","lineage":["https://openalex.org/I14243506"]}],"countries":["HK"],"is_corresponding":false,"raw_author_name":"Bonan Xu","raw_affiliation_strings":["The Hong Kong Polytechnic University"],"affiliations":[{"raw_affiliation_string":"The Hong Kong Polytechnic University","institution_ids":["https://openalex.org/I14243506"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5129641459","display_name":"Shijian Li","orcid":null},"institutions":[{"id":"https://openalex.org/I76130692","display_name":"Zhejiang University","ror":"https://ror.org/00a2xv884","country_code":"CN","type":"education","lineage":["https://openalex.org/I76130692"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Shijian Li","raw_affiliation_strings":["Zhejiang University"],"affiliations":[{"raw_affiliation_string":"Zhejiang University","institution_ids":["https://openalex.org/I76130692"]}]}],"institutions":[],"countries_distinct_count":2,"institutions_distinct_count":3,"corresponding_author_ids":["https://openalex.org/A5129694094"],"corresponding_institution_ids":["https://openalex.org/I76130692"],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":{"value":0.18950749,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":"40","issue":"11","first_page":"9520","last_page":"9528"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.3862000107765198,"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/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.3862000107765198,"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/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":0.13989999890327454,"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/T10036","display_name":"Advanced Neural Network Applications","score":0.13259999454021454,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/transformer","display_name":"Transformer","score":0.5146999955177307},{"id":"https://openalex.org/keywords/feature","display_name":"Feature (linguistics)","score":0.5044000148773193},{"id":"https://openalex.org/keywords/distillation","display_name":"Distillation","score":0.43639999628067017},{"id":"https://openalex.org/keywords/entropy","display_name":"Entropy (arrow of time)","score":0.4300999939441681},{"id":"https://openalex.org/keywords/replicate","display_name":"Replicate","score":0.40380001068115234},{"id":"https://openalex.org/keywords/feature-extraction","display_name":"Feature extraction","score":0.37049999833106995}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6460000276565552},{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.5146999955177307},{"id":"https://openalex.org/C2776401178","wikidata":"https://www.wikidata.org/wiki/Q12050496","display_name":"Feature (linguistics)","level":2,"score":0.5044000148773193},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.49880000948905945},{"id":"https://openalex.org/C204030448","wikidata":"https://www.wikidata.org/wiki/Q101017","display_name":"Distillation","level":2,"score":0.43639999628067017},{"id":"https://openalex.org/C106301342","wikidata":"https://www.wikidata.org/wiki/Q4117933","display_name":"Entropy (arrow of time)","level":2,"score":0.4300999939441681},{"id":"https://openalex.org/C2781162219","wikidata":"https://www.wikidata.org/wiki/Q26250693","display_name":"Replicate","level":2,"score":0.40380001068115234},{"id":"https://openalex.org/C52622490","wikidata":"https://www.wikidata.org/wiki/Q1026626","display_name":"Feature extraction","level":2,"score":0.37049999833106995},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.3334999978542328},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.2831999957561493},{"id":"https://openalex.org/C84945661","wikidata":"https://www.wikidata.org/wiki/Q7366567","display_name":"Root cause","level":2,"score":0.27549999952316284},{"id":"https://openalex.org/C7863114","wikidata":"https://www.wikidata.org/wiki/Q192627","display_name":"Mimicry","level":2,"score":0.2680000066757202},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.2587999999523163},{"id":"https://openalex.org/C8854915","wikidata":"https://www.wikidata.org/wiki/Q4350200","display_name":"Information transfer","level":2,"score":0.2533000111579895}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1609/aaai.v40i11.37913","is_oa":true,"landing_page_url":"https://doi.org/10.1609/aaai.v40i11.37913","pdf_url":null,"source":{"id":"https://openalex.org/S4210191458","display_name":"Proceedings of the AAAI Conference on Artificial Intelligence","issn_l":"2159-5399","issn":["2159-5399","2374-3468"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/P4310320058","host_organization_name":"Association for the Advancement of Artificial Intelligence","host_organization_lineage":["https://openalex.org/P4310320058"],"host_organization_lineage_names":["Association for the Advancement of Artificial Intelligence"],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the AAAI Conference on Artificial Intelligence","raw_type":"journal-article"}],"best_oa_location":{"id":"doi:10.1609/aaai.v40i11.37913","is_oa":true,"landing_page_url":"https://doi.org/10.1609/aaai.v40i11.37913","pdf_url":null,"source":{"id":"https://openalex.org/S4210191458","display_name":"Proceedings of the AAAI Conference on Artificial Intelligence","issn_l":"2159-5399","issn":["2159-5399","2374-3468"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/P4310320058","host_organization_name":"Association for the Advancement of Artificial Intelligence","host_organization_lineage":["https://openalex.org/P4310320058"],"host_organization_lineage_names":["Association for the Advancement of Artificial Intelligence"],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the AAAI Conference on Artificial Intelligence","raw_type":"journal-article"},"sustainable_development_goals":[{"display_name":"Quality Education","id":"https://metadata.un.org/sdg/4","score":0.7476142644882202}],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"While":[0],"feature-based":[1],"knowledge":[2,138],"distillation":[3],"has":[4],"proven":[5],"highly":[6],"effective":[7,162],"for":[8,160],"compressing":[9],"CNNs,":[10],"these":[11,152],"techniques":[12],"unexpectedly":[13],"fail":[14],"when":[15],"applied":[16],"to":[17,118,128,148],"Vision":[18],"Transformers":[19],"(ViTs),":[20],"often":[21],"performing":[22],"worse":[23],"than":[24],"simple":[25],"logit-based":[26],"distillation.":[27],"We":[28,74],"provide":[29],"the":[30,76],"first":[31],"comprehensive":[32],"analysis":[33],"of":[34,79],"this":[35],"phenomenon":[36],"through":[37],"a":[38,63,85],"novel":[39],"analytical":[40],"framework":[41],"termed":[42],"as":[43],"\"distillation":[44],"dynamics\",":[45],"combining":[46],"frequency":[47],"spectrum":[48],"analysis,":[49,97],"information":[50,66],"entropy":[51],"metrics,":[52],"and":[53,92],"activation":[54],"magnitude":[55],"tracking.":[56],"Our":[57,133],"investigation":[58],"reveals":[59],"that":[60,100,111,136,150],"ViTs":[61,141,163],"exhibit":[62],"distinctive":[64],"U-shaped":[65],"processing":[67],"pattern:":[68],"initial":[69],"compression":[70,164],"followed":[71],"by":[72],"expansion.":[73],"identify":[75],"root":[77],"cause":[78],"negative":[80],"transfer":[81,139],"in":[82,108,140],"feature":[83,126,146],"distillation:":[84],"fundamental":[86,153],"representational":[87,154],"paradigm":[88],"mismatch":[89,123],"between":[90],"teacher":[91,101],"student":[93,113,131],"models.":[94],"Through":[95],"frequency-domain":[96],"we":[98],"show":[99],"models":[102,114],"employ":[103],"distributed,":[104],"high-dimensional":[105],"encoding":[106],"strategies":[107],"later":[109],"layers":[110],"smaller":[112],"cannot":[115],"replicate":[116],"due":[117],"limited":[119],"channel":[120],"capacity.":[121],"This":[122],"causes":[124],"late-layer":[125],"alignment":[127],"actively":[129],"harm":[130],"performance.":[132],"findings":[134],"reveal":[135],"successful":[137],"requires":[142],"moving":[143],"beyond":[144],"naive":[145],"mimicry":[147],"methods":[149],"respect":[151],"constraints,":[155],"providing":[156],"essential":[157],"theoretical":[158],"guidance":[159],"designing":[161],"strategies.":[165]},"counts_by_year":[],"updated_date":"2026-03-20T20:47:17.329874","created_date":"2026-03-18T00:00:00"}