{"id":"https://openalex.org/W4417071260","doi":"https://doi.org/10.48550/arxiv.2509.09660","title":"Steering MoE LLMs via Expert (De)Activation","display_name":"Steering MoE LLMs via Expert (De)Activation","publication_year":2025,"publication_date":"2025-09-11","ids":{"openalex":"https://openalex.org/W4417071260","doi":"https://doi.org/10.48550/arxiv.2509.09660"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2509.09660","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2509.09660","pdf_url":"https://arxiv.org/pdf/2509.09660","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,"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-nc-sa","license_id":"https://openalex.org/licenses/cc-by-nc-sa","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"type":"preprint","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2509.09660","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5031986521","display_name":"Mohsen Fayyaz","orcid":"https://orcid.org/0000-0001-5254-3128"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Fayyaz, Mohsen","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100588790","display_name":"Ali Modarressi","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Modarressi, Ali","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5083681053","display_name":"Hanieh Deilamsalehy","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Deilamsalehy, Hanieh","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5028863551","display_name":"Franck Dernoncourt","orcid":"https://orcid.org/0000-0002-1119-1346"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Dernoncourt, Franck","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5120715842","display_name":"Ryan Rossi","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Rossi, Ryan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5033589742","display_name":"Trung Bui","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Bui, Trung","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5071144367","display_name":"Hinrich Sch\u00fctze","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Sch\u00fctze, Hinrich","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5099881261","display_name":"Nanyun Peng","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Peng, Nanyun","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":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":0.5220999717712402,"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":0.5220999717712402,"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":0.08640000224113464,"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/T10883","display_name":"Ethics and Social Impacts of AI","score":0.08160000294446945,"subfield":{"id":"https://openalex.org/subfields/3311","display_name":"Safety Research"},"field":{"id":"https://openalex.org/fields/33","display_name":"Social Sciences"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/key","display_name":"Key (lock)","score":0.6409000158309937},{"id":"https://openalex.org/keywords/control","display_name":"Control (management)","score":0.5332000255584717},{"id":"https://openalex.org/keywords/steering-committee","display_name":"Steering committee","score":0.29820001125335693},{"id":"https://openalex.org/keywords/security-token","display_name":"Security token","score":0.296999990940094},{"id":"https://openalex.org/keywords/poison-control","display_name":"Poison control","score":0.2800000011920929}],"concepts":[{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.6409000158309937},{"id":"https://openalex.org/C2775924081","wikidata":"https://www.wikidata.org/wiki/Q55608371","display_name":"Control (management)","level":2,"score":0.5332000255584717},{"id":"https://openalex.org/C112930515","wikidata":"https://www.wikidata.org/wiki/Q4389547","display_name":"Risk analysis (engineering)","level":1,"score":0.5194000005722046},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.4927000105381012},{"id":"https://openalex.org/C38652104","wikidata":"https://www.wikidata.org/wiki/Q3510521","display_name":"Computer security","level":1,"score":0.4339999854564667},{"id":"https://openalex.org/C127413603","wikidata":"https://www.wikidata.org/wiki/Q11023","display_name":"Engineering","level":0,"score":0.3700999915599823},{"id":"https://openalex.org/C195094911","wikidata":"https://www.wikidata.org/wiki/Q14167904","display_name":"Process management","level":1,"score":0.326200008392334},{"id":"https://openalex.org/C156312663","wikidata":"https://www.wikidata.org/wiki/Q865588","display_name":"Steering committee","level":2,"score":0.29820001125335693},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.296999990940094},{"id":"https://openalex.org/C22212356","wikidata":"https://www.wikidata.org/wiki/Q775325","display_name":"Transport engineering","level":1,"score":0.295199990272522},{"id":"https://openalex.org/C144133560","wikidata":"https://www.wikidata.org/wiki/Q4830453","display_name":"Business","level":0,"score":0.2874999940395355},{"id":"https://openalex.org/C3017944768","wikidata":"https://www.wikidata.org/wiki/Q1450463","display_name":"Poison control","level":2,"score":0.2800000011920929},{"id":"https://openalex.org/C2780791683","wikidata":"https://www.wikidata.org/wiki/Q846785","display_name":"Action (physics)","level":2,"score":0.2720000147819519},{"id":"https://openalex.org/C14036430","wikidata":"https://www.wikidata.org/wiki/Q3736076","display_name":"Function (biology)","level":2,"score":0.2587999999523163},{"id":"https://openalex.org/C132835097","wikidata":"https://www.wikidata.org/wiki/Q7663745","display_name":"System safety","level":2,"score":0.2540000081062317}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2509.09660","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2509.09660","pdf_url":"https://arxiv.org/pdf/2509.09660","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,"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-nc-sa","license_id":"https://openalex.org/licenses/cc-by-nc-sa","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2509.09660","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2509.09660","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,"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":"pmh:oai:arXiv.org:2509.09660","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2509.09660","pdf_url":"https://arxiv.org/pdf/2509.09660","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,"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-nc-sa","license_id":"https://openalex.org/licenses/cc-by-nc-sa","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4417071260.pdf","grobid_xml":"https://content.openalex.org/works/W4417071260.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Mixture-of-Experts":[0],"(MoE)":[1],"in":[2,128],"Large":[3],"Language":[4],"Models":[5],"(LLMs)":[6],"routes":[7],"each":[8],"token":[9],"through":[10],"a":[11,24,116],"subset":[12],"of":[13],"specialized":[14],"Feed-Forward":[15],"Networks":[16],"(FFN),":[17],"known":[18],"as":[19],"experts.":[20,35],"We":[21,36],"present":[22],"SteerMoE,":[23],"framework":[25],"to":[26,87],"steer":[27],"MoE":[28,129],"models":[29],"by":[30,40,85,91,98],"detecting":[31],"and":[32,71,78,89,101,119],"controlling":[33],"behavior-associated":[34],"detect":[37],"key":[38],"experts":[39,63],"comparing":[41],"how":[42],"often":[43],"they":[44],"activate":[45],"between":[46],"paired":[47],"inputs":[48],"that":[49],"demonstrate":[50],"opposite":[51],"behaviors":[52,68],"(e.g.,":[53],"safe":[54],"vs.":[55],"unsafe).":[56],"By":[57],"selectively":[58],"activating":[59],"or":[60],"deactivating":[61],"such":[62],"during":[64],"inference,":[65],"we":[66],"control":[67],"like":[69],"faithfulness":[70,90],"safety":[72,84,97,111],"without":[73],"fine-tuning.":[74],"Across":[75],"11":[76],"benchmarks":[77],"6":[79],"LLMs,":[80],"our":[81],"steering":[82,95],"raises":[83],"up":[86],"+20%":[88],"+27%.":[92],"Alternatively,":[93],"unsafe":[94],"drops":[96],"-41%":[99],"alone,":[100],"-100%":[102],"when":[103],"combined":[104],"with":[105],"existing":[106],"jailbreak":[107],"methods,":[108],"bypassing":[109],"all":[110],"guardrails.":[112],"Overall,":[113],"SteerMoE":[114],"offers":[115],"lightweight,":[117],"effective,":[118],"widely":[120],"applicable":[121],"test-time":[122],"control,":[123],"while":[124],"revealing":[125],"unique":[126],"vulnerabilities":[127],"LLMs.":[130],"https://github.com/adobe-research/SteerMoE":[131]},"counts_by_year":[],"updated_date":"2026-08-24T07:32:12.397491","created_date":"2025-10-10T00:00:00"}
