{"id":"https://openalex.org/W7203728490","doi":"https://doi.org/10.48550/arxiv.2608.15341","title":"TEA: Text Encoder Alignment for Robust Concept Erasure in Text-to-Image Models","display_name":"TEA: Text Encoder Alignment for Robust Concept Erasure in Text-to-Image Models","publication_year":2026,"publication_date":"2026-08-15","ids":{"openalex":"https://openalex.org/W7203728490","doi":"https://doi.org/10.48550/arxiv.2608.15341"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2608.15341","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.15341","pdf_url":"https://arxiv.org/pdf/2608.15341","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2608.15341","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5031956810","display_name":"Alireza dehghanpour farashah","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I4210164802","display_name":"Mila - Quebec Artificial Intelligence Institute","ror":"https://ror.org/05c22rx21","country_code":"CA","type":"facility","lineage":["https://openalex.org/I4210164802"]},{"id":"https://openalex.org/I5023651","display_name":"McGill University","ror":"https://ror.org/01pxwe438","country_code":"CA","type":"education","lineage":["https://openalex.org/I5023651"]}],"countries":["CA"],"is_corresponding":false,"raw_author_name":"Farashah, Alireza Dehghanpour","raw_affiliation_strings":["McGill University","Mila -Quebec AI Institute"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"McGill University","institution_ids":["https://openalex.org/I5023651"]},{"raw_affiliation_string":"Mila -Quebec AI Institute","institution_ids":["https://openalex.org/I4210164802"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5078707812","display_name":"Zhuan Shi","orcid":"https://orcid.org/0000-0003-4239-3546","observed_orcids":["https://orcid.org/0000-0003-4239-3546"]},"institutions":[{"id":"https://openalex.org/I4210164802","display_name":"Mila - Quebec Artificial Intelligence Institute","ror":"https://ror.org/05c22rx21","country_code":"CA","type":"facility","lineage":["https://openalex.org/I4210164802"]},{"id":"https://openalex.org/I5023651","display_name":"McGill University","ror":"https://ror.org/01pxwe438","country_code":"CA","type":"education","lineage":["https://openalex.org/I5023651"]}],"countries":["CA"],"is_corresponding":false,"raw_author_name":"Shi, Zhuan","raw_affiliation_strings":["McGill University","Mila -Quebec AI Institute"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"McGill University","institution_ids":["https://openalex.org/I5023651"]},{"raw_affiliation_string":"Mila -Quebec AI Institute","institution_ids":["https://openalex.org/I4210164802"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5072851389","display_name":"Negar Rostamzadeh","orcid":"https://orcid.org/0000-0002-9638-4664","observed_orcids":["https://orcid.org/0000-0002-9638-4664"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Rostamzadeh, Negar","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5053667504","display_name":"Golnoosh Farnadi","orcid":"https://orcid.org/0000-0003-4504-8668","observed_orcids":["https://orcid.org/0000-0003-4504-8668"]},"institutions":[{"id":"https://openalex.org/I4210164802","display_name":"Mila - Quebec Artificial Intelligence Institute","ror":"https://ror.org/05c22rx21","country_code":"CA","type":"facility","lineage":["https://openalex.org/I4210164802"]},{"id":"https://openalex.org/I5023651","display_name":"McGill University","ror":"https://ror.org/01pxwe438","country_code":"CA","type":"education","lineage":["https://openalex.org/I5023651"]}],"countries":["CA"],"is_corresponding":false,"raw_author_name":"Farnadi, Golnoosh","raw_affiliation_strings":["McGill University","Mila -Quebec AI Institute"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"McGill University","institution_ids":["https://openalex.org/I5023651"]},{"raw_affiliation_string":"Mila -Quebec AI Institute","institution_ids":["https://openalex.org/I4210164802"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":2,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"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":0.9907000064849854,"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":0.9907000064849854,"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/T10775","display_name":"Generative Adversarial Networks and Image Synthesis","score":0.008299999870359898,"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/T12357","display_name":"Digital Media Forensic Detection","score":0.00039999998989515007,"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/concept-erasure","display_name":"concept erasure","score":0.9539999961853027},{"id":"https://openalex.org/keywords/text-to-image-diffusion-models","display_name":"text-to-image diffusion models","score":0.9539999961853027},{"id":"https://openalex.org/keywords/adversarial-prompts","display_name":"adversarial prompts","score":0.7429999709129333},{"id":"https://openalex.org/keywords/domain-alignment","display_name":"domain alignment","score":0.8429999947547913},{"id":"https://openalex.org/keywords/stable-diffusion","display_name":"Stable Diffusion","score":0.8679999709129333},{"id":"https://openalex.org/keywords/black-box-attacks","display_name":"black-box attacks","score":0.6539999842643738},{"id":"https://openalex.org/keywords/white-box-attacks","display_name":"white-box attacks","score":0.906000018119812},{"id":"https://openalex.org/keywords/harmful-content-generation","display_name":"harmful content generation","score":0.6679999828338623}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7954999804496765},{"id":"https://openalex.org/C118505674","wikidata":"https://www.wikidata.org/wiki/Q42586063","display_name":"Encoder","level":2,"score":0.7809000015258789},{"id":"https://openalex.org/C63479239","wikidata":"https://www.wikidata.org/wiki/Q7353546","display_name":"Robustness (evolution)","level":3,"score":0.670199990272522},{"id":"https://openalex.org/C2778790127","wikidata":"https://www.wikidata.org/wiki/Q484885","display_name":"Erasure","level":2,"score":0.6438999772071838},{"id":"https://openalex.org/C137529215","wikidata":"https://www.wikidata.org/wiki/Q5385031","display_name":"Erasure code","level":3,"score":0.4287000000476837},{"id":"https://openalex.org/C2779803651","wikidata":"https://www.wikidata.org/wiki/Q5282088","display_name":"Discriminator","level":3,"score":0.41019999980926514},{"id":"https://openalex.org/C37736160","wikidata":"https://www.wikidata.org/wiki/Q1801315","display_name":"Adversarial system","level":2,"score":0.4049000144004822},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.4041000008583069},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.3515999913215637},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.34060001373291016},{"id":"https://openalex.org/C125411270","wikidata":"https://www.wikidata.org/wiki/Q18653","display_name":"Encoding (memory)","level":2,"score":0.3003000020980835},{"id":"https://openalex.org/C113775141","wikidata":"https://www.wikidata.org/wiki/Q428691","display_name":"Computer engineering","level":1,"score":0.2944999933242798},{"id":"https://openalex.org/C2776359362","wikidata":"https://www.wikidata.org/wiki/Q2145286","display_name":"Representation (politics)","level":3,"score":0.28519999980926514},{"id":"https://openalex.org/C43126263","wikidata":"https://www.wikidata.org/wiki/Q128751","display_name":"Source code","level":2,"score":0.28110000491142273},{"id":"https://openalex.org/C36503486","wikidata":"https://www.wikidata.org/wiki/Q11235244","display_name":"Domain (mathematical analysis)","level":2,"score":0.2806999981403351},{"id":"https://openalex.org/C2779960059","wikidata":"https://www.wikidata.org/wiki/Q7113681","display_name":"Overhead (engineering)","level":2,"score":0.2768000066280365},{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.26930001378059387}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2608.15341","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.15341","pdf_url":"https://arxiv.org/pdf/2608.15341","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2608.15341","doi":"https://doi.org/10.48550/arxiv.2608.15341","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2608.15341","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2608.15341","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.15341","pdf_url":"https://arxiv.org/pdf/2608.15341","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[{"id":"https://openalex.org/F4320309949","display_name":"Canadian Institute for Advanced Research","ror":"https://ror.org/01sdtdd95"},{"id":"https://openalex.org/F4320314000","display_name":"Compute Canada","ror":"https://ror.org/03ty8yr27"},{"id":"https://openalex.org/F4320315389","display_name":"Open Philanthropy Project","ror":"https://ror.org/004d1k391"},{"id":"https://openalex.org/F4320334593","display_name":"Natural Sciences and Engineering Research Council of Canada","ror":"https://ror.org/01h531d29"}],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7203728490.pdf","grobid_xml":"https://content.openalex.org/works/W7203728490.grobid-xml"},"referenced_works_count":5,"referenced_works":["https://openalex.org/W1889081078","https://openalex.org/W4288099666","https://openalex.org/W4305001257","https://openalex.org/W4320342070","https://openalex.org/W7133232659"],"related_works":["https://openalex.org/W4404879643","https://openalex.org/W7155068947","https://openalex.org/W7155062389","https://openalex.org/W4413145310","https://openalex.org/W7202295957","https://openalex.org/W4404198601","https://openalex.org/W7133491908","https://openalex.org/W6979335854","https://openalex.org/W7127418831","https://openalex.org/W6891846030"],"abstract_inverted_index":{"Text-to-image":[0],"diffusion":[1],"models":[2],"can":[3],"be":[4],"misused":[5],"to":[6,96,114,137,184],"generate":[7],"harmful":[8],"content":[9],"through":[10],"adversarial":[11,30,153],"or":[12,37],"paraphrased":[13],"prompts":[14,102],"that":[15,42,74],"bypass":[16],"built-in":[17],"safety":[18],"mechanisms.":[19],"Existing":[20],"concept":[21,53,182],"erasure":[22,54,147,183],"methods":[23,195],"often":[24],"suffer":[25],"from":[26,103],"limited":[27],"robustness":[28,148],"against":[29,149],"prompts,":[31,108],"degradation":[32],"of":[33,100,105,130],"benign":[34,164],"generation":[35,161],"quality,":[36],"reliance":[38],"on":[39,155,163,177],"inference-time":[40,122],"interventions":[41],"introduce":[43],"persistent":[44],"computational":[45],"overhead.":[46],"To":[47],"address":[48],"these":[49,116],"limitations,":[50],"we":[51],"formulate":[52],"as":[55],"a":[56,67,94,127,185],"domain":[57],"alignment":[58],"problem":[59],"in":[60],"the":[61,77,82,111,172],"text":[62,78,112],"representation":[63],"space.":[64],"We":[65],"propose":[66],"lightweight":[68],"Text":[69],"Encoder":[70],"Alignment":[71],"framework":[72],"(TEA)":[73],"fine-tunes":[75],"only":[76,126],"encoder":[79,113],"while":[80,109,159],"keeping":[81],"generative":[83],"backbone":[84],"fully":[85],"frozen.":[86],"Given":[87],"concept--anchor":[88],"prompt":[89],"pairs,":[90],"our":[91],"method":[92],"trains":[93],"discriminator":[95],"distinguish":[97],"token-level":[98],"representations":[99,117],"concept-containing":[101],"those":[104],"safe":[106],"anchor":[107],"updating":[110],"make":[115],"indistinguishable.":[118],"TEA":[119,144,167],"introduces":[120],"zero":[121],"overhead":[123],"and":[124,151,170],"requires":[125],"small":[128],"number":[129],"fine-tuning":[131],"steps,":[132],"making":[133],"it":[134],"highly":[135],"efficient":[136],"deploy":[138],"at":[139,202],"scale.":[140],"Despite":[141],"this":[142],"efficiency,":[143],"achieves":[145,171],"state-of-the-art":[146],"black-box":[150],"white-box":[152],"attacks":[154],"Stable":[156,178],"Diffusion":[157,179],"v1.4,":[158],"preserving":[160],"quality":[162],"prompts.":[165],"Furthermore,":[166],"is":[168,200],"model-agnostic":[169],"lowest":[173],"attack":[174],"success":[175],"rate":[176],"v3.5,":[180],"extending":[181],"Rectified":[186],"Flow":[187],"Transformer":[188],"architecture":[189],"with":[190],"T5":[191],"conditioning":[192],"where":[193],"prior":[194],"remain":[196],"largely":[197],"unexplored.":[198],"Code":[199],"available":[201],"\\href{https://github.com/alirezafarashah/TEA.git}{https://github.com/alirezafarashah/TEA.git}":[203]},"counts_by_year":[],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2026-08-19T00:00:00"}