{"id":"https://openalex.org/W7162639956","doi":"https://doi.org/10.48550/arxiv.2605.28664","title":"Activation Steering for Synthetic Data Generation: The Role of Diversity in Downstream Safety Detection","display_name":"Activation Steering for Synthetic Data Generation: The Role of Diversity in Downstream Safety Detection","publication_year":2026,"publication_date":"2026-05-27","ids":{"openalex":"https://openalex.org/W7162639956","doi":"https://doi.org/10.48550/arxiv.2605.28664"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.28664","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.28664","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":"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"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2605.28664","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5047400791","display_name":"Vijeta Deshpande","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Deshpande, Vijeta","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5074127440","display_name":"Tootiya Giyahchi","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Giyahchi, Tootiya","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137264821","display_name":"Veena Padmanabhan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Padmanabhan, Veena","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137303229","display_name":"Leman Akoglu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Akoglu, Leman","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5071360545","display_name":"Anna Rumshisky","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Rumshisky, Anna","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":0.4253999888896942,"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":0.4253999888896942,"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":0.20640000700950623,"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/T12535","display_name":"Machine Learning and Data Classification","score":0.06340000033378601,"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/downstream","display_name":"Downstream (manufacturing)","score":0.6802999973297119},{"id":"https://openalex.org/keywords/classifier","display_name":"Classifier (UML)","score":0.592199981212616},{"id":"https://openalex.org/keywords/synthetic-data","display_name":"Synthetic data","score":0.5544000267982483},{"id":"https://openalex.org/keywords/training-set","display_name":"Training set","score":0.4634999930858612},{"id":"https://openalex.org/keywords/heuristic","display_name":"Heuristic","score":0.4510999917984009},{"id":"https://openalex.org/keywords/coherence","display_name":"Coherence (philosophical gambling strategy)","score":0.44350001215934753},{"id":"https://openalex.org/keywords/data-quality","display_name":"Data quality","score":0.3889000117778778}],"concepts":[{"id":"https://openalex.org/C2776207758","wikidata":"https://www.wikidata.org/wiki/Q5303302","display_name":"Downstream (manufacturing)","level":2,"score":0.6802999973297119},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6172999739646912},{"id":"https://openalex.org/C95623464","wikidata":"https://www.wikidata.org/wiki/Q1096149","display_name":"Classifier (UML)","level":2,"score":0.592199981212616},{"id":"https://openalex.org/C160920958","wikidata":"https://www.wikidata.org/wiki/Q7662746","display_name":"Synthetic data","level":2,"score":0.5544000267982483},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.4634999930858612},{"id":"https://openalex.org/C173801870","wikidata":"https://www.wikidata.org/wiki/Q201413","display_name":"Heuristic","level":2,"score":0.4510999917984009},{"id":"https://openalex.org/C2781181686","wikidata":"https://www.wikidata.org/wiki/Q4226068","display_name":"Coherence (philosophical gambling strategy)","level":2,"score":0.44350001215934753},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4138999879360199},{"id":"https://openalex.org/C24756922","wikidata":"https://www.wikidata.org/wiki/Q1757694","display_name":"Data quality","level":3,"score":0.3889000117778778},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.3749000132083893},{"id":"https://openalex.org/C111640148","wikidata":"https://www.wikidata.org/wiki/Q847349","display_name":"Rubric","level":2,"score":0.35899999737739563},{"id":"https://openalex.org/C2779530757","wikidata":"https://www.wikidata.org/wiki/Q1207505","display_name":"Quality (philosophy)","level":2,"score":0.3571000099182129},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.3490999937057495},{"id":"https://openalex.org/C2781316041","wikidata":"https://www.wikidata.org/wiki/Q1230584","display_name":"Diversity (politics)","level":2,"score":0.3425999879837036},{"id":"https://openalex.org/C2777488183","wikidata":"https://www.wikidata.org/wiki/Q6900510","display_name":"Safety monitoring","level":2,"score":0.28450000286102295},{"id":"https://openalex.org/C2780977526","wikidata":"https://www.wikidata.org/wiki/Q42417149","display_name":"Data exploration","level":3,"score":0.2766000032424927},{"id":"https://openalex.org/C2781162219","wikidata":"https://www.wikidata.org/wiki/Q26250693","display_name":"Replicate","level":2,"score":0.272599995136261},{"id":"https://openalex.org/C2780440489","wikidata":"https://www.wikidata.org/wiki/Q5227278","display_name":"Data-driven","level":2,"score":0.26840001344680786},{"id":"https://openalex.org/C132964779","wikidata":"https://www.wikidata.org/wiki/Q2110223","display_name":"Raw data","level":2,"score":0.25119999051094055}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.28664","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.28664","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2605.28664","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.28664","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":"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"},"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":{"Safety":[0],"detection":[1,118,206],"models":[2,63],"require":[3],"examples":[4,16,107],"of":[5,72,133,139,164,197],"HHH":[6],"(Helpful,":[7],"Harmless,":[8],"Honest)-violating":[9],"outputs":[10],"for":[11,28,41,186,203,216],"robust":[12],"generalization,":[13],"however":[14],"such":[15],"are":[17],"scarce.":[18],"Activation":[19],"Steering":[20],"(AS)":[21],"has":[22],"emerged":[23],"as":[24,85,210],"a":[25,44,51,86,124,151,182,211],"data-efficient":[26],"method":[27],"generating":[29],"target-concept-aligned":[30],"responses.":[31],"We":[32,49],"investigate":[33],"whether":[34],"AS":[35,141,189,198],"can":[36],"generate":[37],"high-quality":[38],"training":[39,111],"datasets":[40],"downstream":[42,147,170],"classifiers,":[43],"question":[45],"that":[46,96,146,154],"remains":[47],"untested.":[48],"present":[50],"two-fold":[52],"study":[53],"with":[54,113,169],"intrinsic":[55],"and":[56,77,82,94,116,159,178,207],"extrinsic":[57],"evaluation":[58],"across":[59,174],"$4$":[60,134],"concepts":[61,175],"$\\times\\,2$":[62],"$\\times\\,4$":[64],"steering":[65,73,98],"methods.":[66],"Intrinsically,":[67],"beyond":[68],"the":[69,92,109,128,195],"field-standard":[70],"rubric":[71],"success":[74,177],"(concept":[75],"alignment)":[76],"coherence,":[78,158],"we":[79,104],"introduce":[80],"sample-":[81],"set-level":[83],"diversity":[84,209],"quality":[87],"axis":[88,215],"previously":[89,213],"absent":[90],"from":[91],"literature,":[93],"find":[95],"increasing":[97],"strength":[99],"reduces":[100],"response":[101],"diversity.":[102,160],"Extrinsically,":[103],"replace":[105],"HHH-violating":[106],"in":[108,123,150,199],"available":[110],"data":[112,121,130,201],"steered":[114],"generations":[115],"fine-tune":[117],"classifiers.":[119],"AS-generated":[120],"results":[122,193],"better":[125],"classifier":[126],"than":[127,176],"prompting-generated":[129],"on":[131],"$3$":[132],"concepts.":[135],"However,":[136],"only":[137],"$41$":[138],"$136$":[140],"configurations":[142],"outperform":[143],"prompting,":[144],"indicating":[145],"utility":[148],"lies":[149],"narrow":[152],"regime":[153],"jointly":[155],"satisfies":[156],"success,":[157],"The":[161],"harmonic":[162],"mean":[163],"these":[165],"three":[166],"axes":[167],"correlates":[168],"AUROC":[171],"more":[172],"consistently":[173],"coherence":[179],"alone,":[180],"providing":[181],"practical":[183],"heuristic":[184],"target":[185],"practitioners":[187],"tuning":[188,217],"hyperparameters.":[190],"Together,":[191],"our":[192],"highlight":[194],"potential":[196],"synthetic":[200],"generation":[202],"improving":[204],"safety":[205],"identify":[208],"critical,":[212],"overlooked":[214],"AS.":[218]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-05-29T00:00:00"}