{"id":"https://openalex.org/W7148418447","doi":"https://doi.org/10.48550/arxiv.2604.00419","title":"G-Drift MIA: Membership Inference via Gradient-Induced Feature Drift in LLMs","display_name":"G-Drift MIA: Membership Inference via Gradient-Induced Feature Drift in LLMs","publication_year":2026,"publication_date":"2026-04-01","ids":{"openalex":"https://openalex.org/W7148418447","doi":"https://doi.org/10.48550/arxiv.2604.00419"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2604.00419","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.00419","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":"article"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2604.00419","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5132779486","display_name":"Ravi Ranjan","orcid":null},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Ranjan, Ravi","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5132754638","display_name":"Utkarsh Grover","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Grover, Utkarsh","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5132753628","display_name":"Xiaomin Lin","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lin, Xiaomin","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5013726519","display_name":"Agoritsa Polyzou","orcid":"https://orcid.org/0000-0001-8630-7131"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Polyzou, Agoritsa","raw_affiliation_strings":[],"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":4,"corresponding_author_ids":["https://openalex.org/A5132779486"],"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/T10764","display_name":"Privacy-Preserving Technologies in Data","score":0.20579999685287476,"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/T10764","display_name":"Privacy-Preserving Technologies in Data","score":0.20579999685287476,"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/T12262","display_name":"Hate Speech and Cyberbullying Detection","score":0.09229999780654907,"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/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.06790000200271606,"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/inference","display_name":"Inference","score":0.7128000259399414},{"id":"https://openalex.org/keywords/feature","display_name":"Feature (linguistics)","score":0.6384999752044678},{"id":"https://openalex.org/keywords/classifier","display_name":"Classifier (UML)","score":0.5771999955177307},{"id":"https://openalex.org/keywords/concept-drift","display_name":"Concept drift","score":0.5047000050544739},{"id":"https://openalex.org/keywords/representation","display_name":"Representation (politics)","score":0.4772000014781952},{"id":"https://openalex.org/keywords/pattern-recognition","display_name":"Pattern recognition (psychology)","score":0.40209999680519104},{"id":"https://openalex.org/keywords/measure","display_name":"Measure (data warehouse)","score":0.3386000096797943}],"concepts":[{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.7128000259399414},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6430000066757202},{"id":"https://openalex.org/C2776401178","wikidata":"https://www.wikidata.org/wiki/Q12050496","display_name":"Feature (linguistics)","level":2,"score":0.6384999752044678},{"id":"https://openalex.org/C95623464","wikidata":"https://www.wikidata.org/wiki/Q1096149","display_name":"Classifier (UML)","level":2,"score":0.5771999955177307},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5616999864578247},{"id":"https://openalex.org/C60777511","wikidata":"https://www.wikidata.org/wiki/Q3045002","display_name":"Concept drift","level":3,"score":0.5047000050544739},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.49219998717308044},{"id":"https://openalex.org/C2776359362","wikidata":"https://www.wikidata.org/wiki/Q2145286","display_name":"Representation (politics)","level":3,"score":0.4772000014781952},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.460099995136261},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.40209999680519104},{"id":"https://openalex.org/C2780009758","wikidata":"https://www.wikidata.org/wiki/Q6804172","display_name":"Measure (data warehouse)","level":2,"score":0.3386000096797943},{"id":"https://openalex.org/C52622490","wikidata":"https://www.wikidata.org/wiki/Q1026626","display_name":"Feature extraction","level":2,"score":0.33009999990463257},{"id":"https://openalex.org/C148483581","wikidata":"https://www.wikidata.org/wiki/Q446488","display_name":"Feature selection","level":2,"score":0.32690000534057617},{"id":"https://openalex.org/C59404180","wikidata":"https://www.wikidata.org/wiki/Q17013334","display_name":"Feature learning","level":2,"score":0.31690001487731934},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.26170000433921814},{"id":"https://openalex.org/C169258074","wikidata":"https://www.wikidata.org/wiki/Q245748","display_name":"Random forest","level":2,"score":0.2596000134944916},{"id":"https://openalex.org/C83665646","wikidata":"https://www.wikidata.org/wiki/Q42139305","display_name":"Feature vector","level":2,"score":0.2563999891281128},{"id":"https://openalex.org/C116834253","wikidata":"https://www.wikidata.org/wiki/Q2039217","display_name":"Identification (biology)","level":2,"score":0.25429999828338623}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2604.00419","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.00419","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":"article"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2604.00419","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.00419","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":"article"},"sustainable_development_goals":[{"display_name":"Peace, Justice and strong institutions","id":"https://metadata.un.org/sdg/16","score":0.634283721446991}],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Large":[0],"language":[1],"models":[2],"(LLMs)":[3],"are":[4,56,117],"trained":[5],"on":[6,37,72],"massive":[7],"web-scale":[8],"corpora,":[9],"raising":[10],"growing":[11],"concerns":[12],"about":[13],"privacy":[14,200],"and":[15,43,54,91,103,110,135,147,160,176,198],"copyright.":[16],"Membership":[17],"inference":[18,69],"attacks":[19],"(MIAs)":[20],"aim":[21],"to":[22,119],"determine":[23],"whether":[24],"a":[25,66,77,82,121,168,189],"given":[26],"example":[27],"was":[28],"used":[29,118],"during":[30],"training.":[31],"Existing":[32],"LLM":[33],"MIAs":[34],"largely":[35],"rely":[36],"output":[38],"probabilities":[39],"or":[40],"loss":[41,90],"values":[42],"often":[44],"perform":[45],"only":[46],"marginally":[47],"better":[48],"than":[49,165],"random":[50],"guessing":[51],"when":[52],"members":[53,128],"non-members":[55],"drawn":[57],"from":[58,129,138],"the":[59,93,112,194],"same":[60],"distribution.":[61],"We":[62,150],"introduce":[63],"G-Drift":[64,142],"MIA,":[65],"white-box":[67],"membership":[68,195],"method":[70],"based":[71],"gradient-induced":[73],"feature":[74,107,163],"drift.":[75],"Given":[76],"candidate":[78],"(x,y),":[79],"we":[80],"apply":[81],"single":[83],"targeted":[84],"gradient-ascent":[85],"step":[86],"that":[87,125,153,183],"increases":[88],"its":[89],"measure":[92],"resulting":[94],"changes":[95],"in":[96,202],"internal":[97],"representations,":[98],"including":[99],"logits,":[100],"hidden-layer":[101],"activations,":[102],"projections":[104],"onto":[105],"fixed":[106],"directions,":[108],"before":[109],"after":[111],"update.":[113],"These":[114],"drift":[115,164],"signals":[116],"train":[120],"lightweight":[122],"logistic":[123],"classifier":[124],"effectively":[126],"separates":[127],"non-members.":[130],"Across":[131],"multiple":[132],"transformer-based":[133],"LLMs":[134],"datasets":[136],"derived":[137],"realistic":[139],"MIA":[140],"benchmarks,":[141],"substantially":[143],"outperforms":[144],"confidence-based,":[145],"perplexity-based,":[146],"reference-based":[148],"attacks.":[149],"further":[151],"show":[152],"memorized":[154],"training":[155],"samples":[156],"systematically":[157],"exhibit":[158],"smaller":[159],"more":[161],"structured":[162],"non-members,":[166],"providing":[167],"mechanistic":[169],"link":[170],"between":[171],"gradient":[172,186],"geometry,":[173],"representation":[174],"stability,":[175],"memorization.":[177],"In":[178],"general,":[179],"our":[180],"results":[181],"demonstrate":[182],"small,":[184],"controlled":[185],"interventions":[187],"offer":[188],"practical":[190],"tool":[191],"for":[192],"auditing":[193],"of":[196],"training-data":[197],"assessing":[199],"risks":[201],"LLMs.":[203]},"counts_by_year":[],"updated_date":"2026-04-03T16:44:17.987007","created_date":"2026-04-03T00:00:00"}