{"id":"https://openalex.org/W7169680036","doi":"https://doi.org/10.1007/978-981-92-3520-9_13","title":"S3L: Safety Subspace Structured LoRA with Risk-Aware Expert Routing for LLM Safety Alignment","display_name":"S3L: Safety Subspace Structured LoRA with Risk-Aware Expert Routing for LLM Safety Alignment","publication_year":2026,"publication_date":"2026-07-18","ids":{"openalex":"https://openalex.org/W7169680036","doi":"https://doi.org/10.1007/978-981-92-3520-9_13"},"language":"en","primary_location":{"id":"doi:10.1007/978-981-92-3520-9_13","is_oa":true,"landing_page_url":"https://doi.org/10.1007/978-981-92-3520-9_13","pdf_url":"https://link.springer.com/content/pdf/10.1007/978-981-92-3520-9_13.pdf","source":{"id":"https://openalex.org/S106296714","display_name":"Lecture notes in computer science","issn_l":"0302-9743","issn":["0302-9743","1611-3349"],"is_oa":false,"is_in_doaj":false,"is_core":true,"host_organization":"https://openalex.org/P4310319900","host_organization_name":"Springer Science+Business Media","host_organization_lineage":["https://openalex.org/P4310319900","https://openalex.org/P4310319965"],"host_organization_lineage_names":["Springer Science+Business Media","Springer Nature"],"type":"book series"},"license":"cc-by-nc-nd","license_id":"https://openalex.org/licenses/cc-by-nc-nd","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Lecture Notes in Computer Science","raw_type":"book-chapter"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"hybrid","oa_url":"https://link.springer.com/content/pdf/10.1007/978-981-92-3520-9_13.pdf","any_repository_has_fulltext":false},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5008174737","display_name":"Xinghao Wang","orcid":"https://orcid.org/0000-0001-5751-7383"},"institutions":[{"id":"https://openalex.org/I169689159","display_name":"PLA Information Engineering University","ror":"https://ror.org/00mm1qk40","country_code":"CN","type":"education","lineage":["https://openalex.org/I169689159"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Xinghao Wang","raw_affiliation_strings":["Information Engineering University, Zhengzhou, 450001, China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Information Engineering University, Zhengzhou, 450001, China","institution_ids":["https://openalex.org/I169689159"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5111083713","display_name":"Yuke Ma","orcid":"https://orcid.org/0009-0007-2881-6641"},"institutions":[{"id":"https://openalex.org/I169689159","display_name":"PLA Information Engineering University","ror":"https://ror.org/00mm1qk40","country_code":"CN","type":"education","lineage":["https://openalex.org/I169689159"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Yuke Ma","raw_affiliation_strings":["Information Engineering University, Zhengzhou, 450001, China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Information Engineering University, Zhengzhou, 450001, China","institution_ids":["https://openalex.org/I169689159"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5141114681","display_name":"Xi Chen","orcid":null},"institutions":[{"id":"https://openalex.org/I169689159","display_name":"PLA Information Engineering University","ror":"https://ror.org/00mm1qk40","country_code":"CN","type":"education","lineage":["https://openalex.org/I169689159"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Xi Chen","raw_affiliation_strings":["Information Engineering University, Zhengzhou, 450001, China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Information Engineering University, Zhengzhou, 450001, China","institution_ids":["https://openalex.org/I169689159"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5132809870","display_name":"Wei Guo","orcid":null},"institutions":[{"id":"https://openalex.org/I169689159","display_name":"PLA Information Engineering University","ror":"https://ror.org/00mm1qk40","country_code":"CN","type":"education","lineage":["https://openalex.org/I169689159"]}],"countries":["CN"],"is_corresponding":true,"raw_author_name":"Wei Guo","raw_affiliation_strings":["Information Engineering University, Zhengzhou, 450001, China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Information Engineering University, Zhengzhou, 450001, China","institution_ids":["https://openalex.org/I169689159"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":["https://openalex.org/A5132809870"],"corresponding_institution_ids":["https://openalex.org/I169689159"],"apc_list":{"value":5000,"currency":"EUR","value_usd":5392},"apc_paid":{"value":5000,"currency":"EUR","value_usd":5392},"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":"157","last_page":"169"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":null,"topics":[],"keywords":[{"id":"https://openalex.org/keywords/subspace-topology","display_name":"Subspace topology","score":0.7538999915122986},{"id":"https://openalex.org/keywords/discriminative-model","display_name":"Discriminative model","score":0.734000027179718},{"id":"https://openalex.org/keywords/adapter","display_name":"Adapter (computing)","score":0.6406999826431274},{"id":"https://openalex.org/keywords/representation","display_name":"Representation (politics)","score":0.4708999991416931},{"id":"https://openalex.org/keywords/train","display_name":"Train","score":0.4587000012397766},{"id":"https://openalex.org/keywords/safer","display_name":"SAFER","score":0.4429999887943268},{"id":"https://openalex.org/keywords/calibration","display_name":"Calibration","score":0.3986000120639801}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.901199996471405},{"id":"https://openalex.org/C32834561","wikidata":"https://www.wikidata.org/wiki/Q660730","display_name":"Subspace topology","level":2,"score":0.7538999915122986},{"id":"https://openalex.org/C97931131","wikidata":"https://www.wikidata.org/wiki/Q5282087","display_name":"Discriminative model","level":2,"score":0.734000027179718},{"id":"https://openalex.org/C177284502","wikidata":"https://www.wikidata.org/wiki/Q1005390","display_name":"Adapter (computing)","level":2,"score":0.6406999826431274},{"id":"https://openalex.org/C2776359362","wikidata":"https://www.wikidata.org/wiki/Q2145286","display_name":"Representation (politics)","level":3,"score":0.4708999991416931},{"id":"https://openalex.org/C190839683","wikidata":"https://www.wikidata.org/wiki/Q2448197","display_name":"Train","level":2,"score":0.4587000012397766},{"id":"https://openalex.org/C2776654903","wikidata":"https://www.wikidata.org/wiki/Q2601463","display_name":"SAFER","level":2,"score":0.4429999887943268},{"id":"https://openalex.org/C79403827","wikidata":"https://www.wikidata.org/wiki/Q3988","display_name":"Real-time computing","level":1,"score":0.40860000252723694},{"id":"https://openalex.org/C165838908","wikidata":"https://www.wikidata.org/wiki/Q736777","display_name":"Calibration","level":2,"score":0.3986000120639801},{"id":"https://openalex.org/C74172769","wikidata":"https://www.wikidata.org/wiki/Q1446839","display_name":"Routing (electronic design automation)","level":2,"score":0.3425000011920929},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.34200000762939453},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.32820001244544983},{"id":"https://openalex.org/C2777488183","wikidata":"https://www.wikidata.org/wiki/Q6900510","display_name":"Safety monitoring","level":2,"score":0.3224000036716461},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.32170000672340393},{"id":"https://openalex.org/C2779843651","wikidata":"https://www.wikidata.org/wiki/Q7390335","display_name":"SIGNAL (programming language)","level":2,"score":0.3151000142097473},{"id":"https://openalex.org/C2776502983","wikidata":"https://www.wikidata.org/wiki/Q690182","display_name":"Contrast (vision)","level":2,"score":0.28529998660087585},{"id":"https://openalex.org/C2986542766","wikidata":"https://www.wikidata.org/wiki/Q2090494","display_name":"Vehicle safety","level":2,"score":0.28189998865127563},{"id":"https://openalex.org/C2776240099","wikidata":"https://www.wikidata.org/wiki/Q327018","display_name":"Interrogation","level":2,"score":0.27230000495910645},{"id":"https://openalex.org/C50942859","wikidata":"https://www.wikidata.org/wiki/Q4967193","display_name":"Rectification","level":3,"score":0.27000001072883606},{"id":"https://openalex.org/C149635348","wikidata":"https://www.wikidata.org/wiki/Q193040","display_name":"Embedded system","level":1,"score":0.26570001244544983}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1007/978-981-92-3520-9_13","is_oa":true,"landing_page_url":"https://doi.org/10.1007/978-981-92-3520-9_13","pdf_url":"https://link.springer.com/content/pdf/10.1007/978-981-92-3520-9_13.pdf","source":{"id":"https://openalex.org/S106296714","display_name":"Lecture notes in computer science","issn_l":"0302-9743","issn":["0302-9743","1611-3349"],"is_oa":false,"is_in_doaj":false,"is_core":true,"host_organization":"https://openalex.org/P4310319900","host_organization_name":"Springer Science+Business Media","host_organization_lineage":["https://openalex.org/P4310319900","https://openalex.org/P4310319965"],"host_organization_lineage_names":["Springer Science+Business Media","Springer Nature"],"type":"book series"},"license":"cc-by-nc-nd","license_id":"https://openalex.org/licenses/cc-by-nc-nd","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Lecture Notes in Computer Science","raw_type":"book-chapter"}],"best_oa_location":{"id":"doi:10.1007/978-981-92-3520-9_13","is_oa":true,"landing_page_url":"https://doi.org/10.1007/978-981-92-3520-9_13","pdf_url":"https://link.springer.com/content/pdf/10.1007/978-981-92-3520-9_13.pdf","source":{"id":"https://openalex.org/S106296714","display_name":"Lecture notes in computer science","issn_l":"0302-9743","issn":["0302-9743","1611-3349"],"is_oa":false,"is_in_doaj":false,"is_core":true,"host_organization":"https://openalex.org/P4310319900","host_organization_name":"Springer Science+Business Media","host_organization_lineage":["https://openalex.org/P4310319900","https://openalex.org/P4310319965"],"host_organization_lineage_names":["Springer Science+Business Media","Springer Nature"],"type":"book series"},"license":"cc-by-nc-nd","license_id":"https://openalex.org/licenses/cc-by-nc-nd","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Lecture Notes in Computer Science","raw_type":"book-chapter"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7169680036.pdf","grobid_xml":"https://content.openalex.org/works/W7169680036.grobid-xml"},"referenced_works_count":5,"referenced_works":["https://openalex.org/W2805695370","https://openalex.org/W4401042900","https://openalex.org/W4402671841","https://openalex.org/W4415795355","https://openalex.org/W4415797606"],"related_works":[],"abstract_inverted_index":{"Safety":[0,46],"tuning":[1,187],"for":[2,178,183],"large":[3],"language":[4],"models":[5],"(LLMs)":[6],"has":[7],"to":[8,117,189],"balance":[9],"two":[10],"errors:":[11],"complying":[12],"with":[13,88,156],"a":[14,19,32,51,59,65,170],"jailbreak":[15,192],"request":[16],"and":[17,77,94,107,140,161],"refusing":[18],"legitimate":[20],"but":[21,181],"sensitive":[22],"query.":[23],"We":[24,44],"study":[25],"this":[26],"trade-off":[27],"under":[28,191],"LoRA":[29,49,53,86,136],"adaptation,":[30],"where":[31],"compact":[33],"update":[34],"can":[35],"noticeably":[36],"change":[37],"the":[38,105,115,144],"refusal":[39],"behavior":[40],"inherited":[41],"from":[42],"alignment.":[43],"present":[45],"Subspace":[47],"Structured":[48],"(S3L),":[50],"multi-expert":[52],"framework":[54],"that":[55,74,169],"selects":[56],"adapters":[57],"using":[58],"calibrated":[60],"representation":[61],"signal":[62],"rather":[63],"than":[64],"fixed":[66],"adapter":[67],"average.":[68],"S3L":[69,132],"first":[70],"estimates":[71],"cPCA":[72],"directions":[73],"contrast":[75],"harmful":[76,154],"benign":[78],"calibration":[79],"activations.":[80],"It":[81],"then":[82],"trains":[83],"three":[84],"late-layer":[85],"experts":[87],"different":[89],"safety-utility":[90],"biases,":[91],"auxiliary":[92],"regularization,":[93],"update-direction":[95],"separation.":[96],"During":[97],"inference,":[98],"frozen":[99],"input":[100],"activations":[101],"are":[102],"projected":[103],"into":[104],"safety":[106,172,186],"hard-safe":[108],"subspaces.":[109],"A":[110],"two-stage":[111],"boundary-band":[112],"rule":[113],"assigns":[114],"prompt":[116],"exactly":[118],"one":[119],"expert,":[120],"so":[121],"generation":[122],"uses":[123],"no":[124],"weight,":[125],"logit,":[126],"or":[127],"hidden-state":[128],"fusion.":[129],"On":[130,143],"Zephyr-7B-Beta,":[131],"improves":[133],"over":[134],"static":[135],"experts,":[137],"weight":[138],"averaging,":[139],"one-stage":[141],"routing.":[142],"full":[145],"evaluation":[146],"sets,":[147],"it":[148],"obtains":[149],"5.85%":[150],"ASR":[151],"on":[152,159,164],"188":[153],"prompts,":[155],"8.00%":[157],"ORR":[158,163],"B-Core":[160],"9.50%":[162],"ORBench.":[165],"The":[166],"results":[167],"suggest":[168],"discriminative":[171],"subspace":[173],"is":[174],"useful":[175],"not":[176],"only":[177],"scoring":[179],"risk,":[180],"also":[182],"making":[184],"multi-adapter":[185],"easier":[188],"inspect":[190],"evaluation.":[193]},"counts_by_year":[],"updated_date":"2026-07-20T05:51:44.042241","created_date":"2026-07-19T00:00:00"}