{"id":"https://openalex.org/W7138353352","doi":"https://doi.org/10.48550/arxiv.2603.13670","title":"SecDTD: Dynamic Token Drop for Secure Transformers Inference","display_name":"SecDTD: Dynamic Token Drop for Secure Transformers Inference","publication_year":2026,"publication_date":"2026-03-14","ids":{"openalex":"https://openalex.org/W7138353352","doi":"https://doi.org/10.48550/arxiv.2603.13670"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2603.13670","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.13670","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2603.13670","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5068070082","display_name":"Yifei Cai","orcid":"https://orcid.org/0000-0002-1372-656X"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Cai, Yifei","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100689266","display_name":"Zhuoran Li","orcid":"https://orcid.org/0000-0001-6281-2248"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Zhuoran","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5091764453","display_name":"Yizhou Feng","orcid":"https://orcid.org/0000-0002-3520-7015"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Feng, Yizhou","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5129648791","display_name":"Qiao Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Qiao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5129719980","display_name":"Hongyi Wu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wu, Hongyi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5078136344","display_name":"Danella Zhao","orcid":"https://orcid.org/0009-0007-5688-8466"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhao, Danella","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5074327315","display_name":"Chunsheng Xin","orcid":"https://orcid.org/0000-0001-5575-2849"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xin, Chunsheng","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/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.5224999785423279,"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":0.5224999785423279,"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/T10764","display_name":"Privacy-Preserving Technologies in Data","score":0.20340000092983246,"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/T10237","display_name":"Cryptography and Data Security","score":0.14169999957084656,"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/security-token","display_name":"Security token","score":0.791100025177002},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.6459000110626221},{"id":"https://openalex.org/keywords/speedup","display_name":"Speedup","score":0.4952000081539154},{"id":"https://openalex.org/keywords/plaintext","display_name":"Plaintext","score":0.4246000051498413},{"id":"https://openalex.org/keywords/token-passing","display_name":"Token passing","score":0.41990000009536743},{"id":"https://openalex.org/keywords/overhead","display_name":"Overhead (engineering)","score":0.375900000333786},{"id":"https://openalex.org/keywords/drop","display_name":"Drop (telecommunication)","score":0.3734999895095825},{"id":"https://openalex.org/keywords/paillier-cryptosystem","display_name":"Paillier cryptosystem","score":0.3449000120162964}],"concepts":[{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.791100025177002},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7858999967575073},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.6459000110626221},{"id":"https://openalex.org/C68339613","wikidata":"https://www.wikidata.org/wiki/Q1549489","display_name":"Speedup","level":2,"score":0.4952000081539154},{"id":"https://openalex.org/C31258907","wikidata":"https://www.wikidata.org/wiki/Q1301371","display_name":"Computer network","level":1,"score":0.4699999988079071},{"id":"https://openalex.org/C120314980","wikidata":"https://www.wikidata.org/wiki/Q180634","display_name":"Distributed computing","level":1,"score":0.45590001344680786},{"id":"https://openalex.org/C92717368","wikidata":"https://www.wikidata.org/wiki/Q1162538","display_name":"Plaintext","level":3,"score":0.4246000051498413},{"id":"https://openalex.org/C115067241","wikidata":"https://www.wikidata.org/wiki/Q1639854","display_name":"Token passing","level":3,"score":0.41990000009536743},{"id":"https://openalex.org/C2779960059","wikidata":"https://www.wikidata.org/wiki/Q7113681","display_name":"Overhead (engineering)","level":2,"score":0.375900000333786},{"id":"https://openalex.org/C2781345722","wikidata":"https://www.wikidata.org/wiki/Q5308388","display_name":"Drop (telecommunication)","level":2,"score":0.3734999895095825},{"id":"https://openalex.org/C66989864","wikidata":"https://www.wikidata.org/wiki/Q594646","display_name":"Paillier cryptosystem","level":5,"score":0.3449000120162964},{"id":"https://openalex.org/C2780513914","wikidata":"https://www.wikidata.org/wiki/Q18210350","display_name":"Bottleneck","level":2,"score":0.3303000032901764},{"id":"https://openalex.org/C79403827","wikidata":"https://www.wikidata.org/wiki/Q3988","display_name":"Real-time computing","level":1,"score":0.3147999942302704},{"id":"https://openalex.org/C93974786","wikidata":"https://www.wikidata.org/wiki/Q1589480","display_name":"Ciphertext","level":3,"score":0.3034000098705292},{"id":"https://openalex.org/C136886441","wikidata":"https://www.wikidata.org/wiki/Q926129","display_name":"Normalization (sociology)","level":2,"score":0.29589998722076416},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.28769999742507935},{"id":"https://openalex.org/C178489894","wikidata":"https://www.wikidata.org/wiki/Q8789","display_name":"Cryptography","level":2,"score":0.28200000524520874},{"id":"https://openalex.org/C93996380","wikidata":"https://www.wikidata.org/wiki/Q44127","display_name":"Server","level":2,"score":0.2784000039100647},{"id":"https://openalex.org/C160234255","wikidata":"https://www.wikidata.org/wiki/Q812535","display_name":"Bayesian inference","level":3,"score":0.273499995470047},{"id":"https://openalex.org/C113775141","wikidata":"https://www.wikidata.org/wiki/Q428691","display_name":"Computer engineering","level":1,"score":0.27219998836517334},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.27149999141693115},{"id":"https://openalex.org/C175291020","wikidata":"https://www.wikidata.org/wiki/Q1156822","display_name":"Offset (computer science)","level":2,"score":0.2685999870300293},{"id":"https://openalex.org/C2780385302","wikidata":"https://www.wikidata.org/wiki/Q367158","display_name":"Protocol (science)","level":3,"score":0.26339998841285706},{"id":"https://openalex.org/C57273362","wikidata":"https://www.wikidata.org/wiki/Q576722","display_name":"Decoding methods","level":2,"score":0.26179999113082886},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.25}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2603.13670","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.13670","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2603.13670","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.13670","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"The":[0],"rapid":[1],"adoption":[2],"of":[3,125,177],"Transformer-based":[4],"AI":[5],"has":[6],"been":[7,48],"driven":[8],"by":[9,113],"accessible":[10],"models":[11],"such":[12,81,128],"as":[13,24,37,129],"ChatGPT,":[14],"which":[15],"provide":[16],"API-based":[17],"services":[18,28],"for":[19,70,105],"developers":[20],"and":[21,54,155,198,217],"businesses.":[22],"However,":[23],"these":[25],"online":[26],"inference":[27,45,72,119,225],"increasingly":[29],"handle":[30],"sensitive":[31],"inputs,":[32],"privacy":[33],"concerns":[34],"have":[35,47],"emerged":[36],"a":[38,99,191],"significant":[39],"challenge.":[40],"To":[41,131],"address":[42],"this,":[43,133],"secure":[44,94,106],"frameworks":[46],"proposed,":[49],"but":[50],"their":[51],"high":[52],"computational":[53],"communication":[55],"overhead":[56,154],"often":[57],"limit":[58],"practical":[59],"deployment.":[60],"In":[61],"plaintext":[62],"settings,":[63],"token":[64,101,111,150,182],"drop":[65,102,112,151],"is":[66,86],"an":[67],"effective":[68],"technique":[69],"reducing":[71,122],"cost;":[73],"however,":[74],"our":[75],"analysis":[76],"reveals":[77],"that":[78,147,172],"directly":[79],"applying":[80],"methods":[82],"to":[83,89,117,180,185],"ciphertext":[84],"scenarios":[85],"suboptimal":[87],"due":[88],"distinct":[90],"cost":[91,124],"distributions":[92],"in":[93,229],"computation.":[95],"We":[96,200],"propose":[97],"SecDTD,":[98],"dynamic":[100],"scheme":[103],"tailored":[104],"Transformer":[107],"inference.":[108],"SecDTD":[109,202,220],"advances":[110],"shifting":[114],"the":[115,123,175,215],"dropping":[116,161],"earlier":[118],"stages,":[120],"effectively":[121],"key":[126],"components":[127],"Softmax.":[130],"support":[132,181],"we":[134],"introduce":[135],"two":[136],"core":[137],"techniques.":[138],"Max-Centric":[139],"Normalization":[140],"(MCN):":[141],"A":[142,166],"novel,":[143],"Softmax-independent":[144],"scoring":[145],"method":[146],"enables":[148],"early":[149],"with":[152],"minimal":[153],"improved":[156],"normalization,":[157],"supporting":[158],"more":[159],"aggressive":[160],"without":[162,227],"accuracy":[163],"loss.":[164],"OMSel:":[165],"faster,":[167],"oblivious":[168],"median":[169,176],"selection":[170],"protocol":[171],"securely":[173],"identifies":[174],"importance":[178],"scores":[179],"drop.":[183],"Compared":[184],"existing":[186],"sorting-based":[187],"methods,":[188],"OMSel":[189],"achieves":[190,221],"16.9$\\times$":[192],"speedup":[193],"while":[194],"maintaining":[195],"security,":[196],"obliviousness":[197],"randomness.":[199],"evaluate":[201],"through":[203],"48":[204],"experiments":[205],"across":[206],"eight":[207],"GLUE":[208],"datasets":[209],"under":[210],"various":[211],"network":[212],"settings":[213],"using":[214],"BOLT":[216],"BumbleBee":[218],"frameworks.":[219],"4.47":[222],"times":[223],"end-to-end":[224],"acceleration":[226],"degradation":[228],"accuracy.":[230]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-03-18T00:00:00"}