{"id":"https://openalex.org/W7225259694","doi":"https://doi.org/10.48550/arxiv.2611.04657","title":"Exploiting Biases in Attention for Transformers","display_name":"Exploiting Biases in Attention for Transformers","publication_year":2026,"publication_date":"2026-11-03","ids":{"openalex":"https://openalex.org/W7225259694","doi":"https://doi.org/10.48550/arxiv.2611.04657"},"language":"en","primary_location":{"id":"doi:10.48550/arxiv.2611.04657","doi":"https://doi.org/10.48550/arxiv.2611.04657","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2611.04657","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2611.04657","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5160644308","display_name":"Charles Herr","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Herr, Charles","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5092741233","display_name":"Gabriele Spadaro","orcid":"https://orcid.org/0009-0008-4786-1074","observed_orcids":["https://orcid.org/0009-0008-4786-1074"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Spadaro, Gabriele","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5046851981","display_name":"Enzo Tartaglione","orcid":"https://orcid.org/0000-0003-4274-8298","observed_orcids":["https://orcid.org/0000-0003-4274-8298"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Tartaglione, Enzo","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":false,"cited_by_count":0,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10036","display_name":"Advanced Neural Network Applications","score":0.9907000064849854,"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/T10036","display_name":"Advanced Neural Network Applications","score":0.9907000064849854,"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/T11605","display_name":"Visual Attention and Saliency Detection","score":0.006500000134110451,"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":0.0005000000237487257,"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/vision-transformer","display_name":"vision transformer","score":0.8130000233650208},{"id":"https://openalex.org/keywords/sparse-attention","display_name":"sparse attention","score":0.8899999856948853},{"id":"https://openalex.org/keywords/attentional-bias","display_name":"attentional bias","score":0.652999997138977},{"id":"https://openalex.org/keywords/token-importance","display_name":"token importance","score":0.6729999780654907},{"id":"https://openalex.org/keywords/computational-cost-reduction","display_name":"computational complexity reduction","score":0.6949999928474426},{"id":"https://openalex.org/keywords/inference-latency","display_name":"inference latency","score":0.9520000219345093}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6736999750137329},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.6399999856948853},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.6258000135421753},{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.6075999736785889},{"id":"https://openalex.org/C50335755","wikidata":"https://www.wikidata.org/wiki/Q483247","display_name":"Phenomenon","level":2,"score":0.5436999797821045},{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.4726000130176544},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.39640000462532043},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3962000012397766},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.3853999972343445},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.3781000077724457},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.3391999900341034},{"id":"https://openalex.org/C108010975","wikidata":"https://www.wikidata.org/wiki/Q500094","display_name":"Pruning","level":2,"score":0.3366999924182892},{"id":"https://openalex.org/C158154518","wikidata":"https://www.wikidata.org/wiki/Q7310970","display_name":"Relevance (law)","level":2,"score":0.33320000767707825},{"id":"https://openalex.org/C57691317","wikidata":"https://www.wikidata.org/wiki/Q1289248","display_name":"Scalar (mathematics)","level":2,"score":0.2831000089645386},{"id":"https://openalex.org/C2780148112","wikidata":"https://www.wikidata.org/wiki/Q1432581","display_name":"Proxy (statistics)","level":2,"score":0.2759000062942505},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.26759999990463257},{"id":"https://openalex.org/C2776291640","wikidata":"https://www.wikidata.org/wiki/Q2912517","display_name":"Value (mathematics)","level":2,"score":0.2655999958515167},{"id":"https://openalex.org/C3826847","wikidata":"https://www.wikidata.org/wiki/Q188768","display_name":"FLOPS","level":2,"score":0.262800008058548}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2611.04657","doi":"https://doi.org/10.48550/arxiv.2611.04657","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2611.04657","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2611.04657","doi":"https://doi.org/10.48550/arxiv.2611.04657","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2611.04657","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4417300299","https://openalex.org/W4372266714","https://openalex.org/W3173222967","https://openalex.org/W7127166091","https://openalex.org/W4405783272","https://openalex.org/W7161304091","https://openalex.org/W4417093983","https://openalex.org/W4403884248","https://openalex.org/W4407570359","https://openalex.org/W4372266912"],"abstract_inverted_index":{"Attention":[0],"is":[1],"known":[2],"to":[3,19,74,121],"be":[4,36],"sparse":[5],"in":[6,129],"vision":[7],"transformers.":[8],"In":[9],"particular,":[10],"we":[11,56],"analyze":[12],"the":[13,20,47,61,75,99,124,141,148],"unequal":[14],"relevance":[15],"of":[16,51,60,127,131,140],"attention-matrix":[17],"columns":[18],"final":[21],"computation":[22],"and":[23,102,134,150],"show":[24,80],"that":[25,40,64,81],"this":[26,66,82,118],"phenomenon":[27,67],"becomes":[28],"increasingly":[29],"pronounced":[30],"with":[31],"network":[32],"depth.":[33],"This":[34],"can":[35,103],"interpreted":[37],"as":[38,106],"saying":[39],"tokens":[41,97],"have":[42],"an":[43,71],"absolute":[44],"importance":[45],"for":[46,109],"attention":[48,62,100,111],"process,":[49,101],"independently":[50],"queries.":[52],"Inspired":[53],"by":[54,69],"this,":[55],"propose":[57],"a":[58,88,107],"reformulation":[59],"formula":[63],"makes":[65],"explicit":[68],"adding":[70],"input-dependent":[72],"bias":[73],"query-key":[76],"dot":[77],"product.":[78],"We":[79,145],"new":[83],"term":[84],"gives":[85],"each":[86],"token":[87],"scalar":[89],"value":[90],"indicating":[91],"its":[92],"anticipated":[93],"influence":[94],"on":[95],"other":[96],"during":[98],"thus":[104],"serve":[105],"proxy":[108],"pruning":[110],"maps.":[112],"Across":[113],"multiple":[114],"transformer":[115],"model":[116],"sizes,":[117],"allows":[119],"us":[120],"substantially":[122],"reduce":[123],"computational":[125],"cost":[126],"attention,":[128],"terms":[130],"both":[132],"FLOPs":[133],"inference":[135],"latency,":[136],"while":[137],"retaining":[138],"most":[139],"original":[142],"classification":[143],"accuracy.":[144],"publicly":[146],"released":[147],"code":[149],"models":[151],"at":[152],"https://github.com/EIDOSLAB/AttentionWithBias.":[153]},"counts_by_year":[],"updated_date":"2026-11-06T12:05:46.655746","created_date":"2026-11-05T00:00:00"}