{"id":"https://openalex.org/W4412944574","doi":"https://doi.org/10.18653/v1/2025.findings-acl.1308","title":"CogSteer: Cognition-Inspired Selective Layer Intervention for Efficiently Steering Large Language Models","display_name":"CogSteer: Cognition-Inspired Selective Layer Intervention for Efficiently Steering Large Language Models","publication_year":2025,"publication_date":"2025-01-01","ids":{"openalex":"https://openalex.org/W4412944574","doi":"https://doi.org/10.18653/v1/2025.findings-acl.1308"},"language":"en","primary_location":{"id":"doi:10.18653/v1/2025.findings-acl.1308","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-acl.1308","pdf_url":"https://aclanthology.org/2025.findings-acl.1308.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: ACL 2025","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2025.findings-acl.1308.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5100783586","display_name":"Xintong Wang","orcid":"https://orcid.org/0009-0007-5208-2609"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xintong Wang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5111333819","display_name":"Jingheng Pan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jingheng Pan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5027478572","display_name":"Liang Ding","orcid":"https://orcid.org/0000-0003-3998-3860"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liang Ding","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5088191810","display_name":"Longyue Wang","orcid":"https://orcid.org/0000-0002-9062-6183"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Longyue Wang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5078191591","display_name":"Longqin Jiang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Longqin Jiang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5007862457","display_name":"Xingshan Li","orcid":"https://orcid.org/0000-0002-0995-4631"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xingshan Li","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5021287757","display_name":"Chris Biemann","orcid":"https://orcid.org/0000-0002-8449-9624"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chris Biemann","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":"25507","last_page":"25522"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.9541000127792358,"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/T10028","display_name":"Topic Modeling","score":0.9541000127792358,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.9212999939918518,"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/computer-science","display_name":"Computer science","score":0.6711153388023376},{"id":"https://openalex.org/keywords/cognition","display_name":"Cognition","score":0.6294065117835999},{"id":"https://openalex.org/keywords/layer","display_name":"Layer (electronics)","score":0.6212997436523438},{"id":"https://openalex.org/keywords/intervention","display_name":"Intervention (counseling)","score":0.44580283761024475},{"id":"https://openalex.org/keywords/psychology","display_name":"Psychology","score":0.20668253302574158},{"id":"https://openalex.org/keywords/neuroscience","display_name":"Neuroscience","score":0.11965182423591614},{"id":"https://openalex.org/keywords/materials-science","display_name":"Materials science","score":0.08500117063522339},{"id":"https://openalex.org/keywords/nanotechnology","display_name":"Nanotechnology","score":0.06557422876358032}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6711153388023376},{"id":"https://openalex.org/C169900460","wikidata":"https://www.wikidata.org/wiki/Q2200417","display_name":"Cognition","level":2,"score":0.6294065117835999},{"id":"https://openalex.org/C2779227376","wikidata":"https://www.wikidata.org/wiki/Q6505497","display_name":"Layer (electronics)","level":2,"score":0.6212997436523438},{"id":"https://openalex.org/C2780665704","wikidata":"https://www.wikidata.org/wiki/Q959298","display_name":"Intervention (counseling)","level":2,"score":0.44580283761024475},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.20668253302574158},{"id":"https://openalex.org/C169760540","wikidata":"https://www.wikidata.org/wiki/Q207011","display_name":"Neuroscience","level":1,"score":0.11965182423591614},{"id":"https://openalex.org/C192562407","wikidata":"https://www.wikidata.org/wiki/Q228736","display_name":"Materials science","level":0,"score":0.08500117063522339},{"id":"https://openalex.org/C171250308","wikidata":"https://www.wikidata.org/wiki/Q11468","display_name":"Nanotechnology","level":1,"score":0.06557422876358032},{"id":"https://openalex.org/C118552586","wikidata":"https://www.wikidata.org/wiki/Q7867","display_name":"Psychiatry","level":1,"score":0.0}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2025.findings-acl.1308","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-acl.1308","pdf_url":"https://aclanthology.org/2025.findings-acl.1308.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: ACL 2025","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2025.findings-acl.1308","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-acl.1308","pdf_url":"https://aclanthology.org/2025.findings-acl.1308.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: ACL 2025","raw_type":"proceedings-article"},"sustainable_development_goals":[{"score":0.41999998688697815,"display_name":"Quality Education","id":"https://metadata.un.org/sdg/4"}],"awards":[],"funders":[{"id":"https://openalex.org/F4320323381","display_name":"Universit\u00e4t Hamburg","ror":"https://ror.org/00g30e956"}],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4412944574.pdf","grobid_xml":"https://content.openalex.org/works/W4412944574.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4391375266","https://openalex.org/W2899084033","https://openalex.org/W2748952813","https://openalex.org/W2390279801","https://openalex.org/W4391913857","https://openalex.org/W2358668433","https://openalex.org/W4396701345","https://openalex.org/W2376932109","https://openalex.org/W2001405890","https://openalex.org/W4396696052"],"abstract_inverted_index":{"Large":[0],"Language":[1],"Models":[2],"(LLMs)":[3],"achieve":[4],"remarkable":[5],"performance":[6],"through":[7],"pretraining":[8],"on":[9,66,94,127,136],"extensive":[10],"data.This":[11],"enables":[12],"efficient":[13,89],"adaptation":[14],"to":[15,30,80,119],"diverse":[16],"downstream":[17],"tasks.However,":[18],"the":[19,28,41,56,76,107,142,154],"lack":[20],"of":[21,44,146,156],"interpretability":[22,155],"in":[23],"their":[24],"underlying":[25],"mechanisms":[26,43],"limits":[27],"ability":[29],"effectively":[31],"steer":[32,120],"LLMs":[33,45,121,157],"for":[34,74,161],"specific":[35],"applications.In":[36],"this":[37,84],"work,":[38],"we":[39,54,69,86,110],"investigate":[40],"intrinsic":[42],"from":[46,123],"a":[47,71,149],"cognitive":[48,61],"perspective":[49],"using":[50],"eye":[51],"movement":[52],"measures.Specifically,":[53],"analyze":[55],"layer-wise":[57],"correlation":[58],"between":[59],"human":[60],"indicators":[62],"and":[63,132,139,144],"LLM":[64,82],"representations.Building":[65],"these":[67],"insights,":[68],"propose":[70],"heuristic":[72],"approach":[73],"selecting":[75],"optimal":[77],"steering":[78],"layer":[79,91,114],"modulate":[81],"semantics.To":[83],"end,":[85],"introduce":[87],"an":[88,112],"selective":[90],"intervention":[92,116],"based":[93],"prominent":[95],"parameter-efficient":[96],"fine-tuning":[97],"methods,":[98],"which":[99],"conventionally":[100],"adjust":[101],"either":[102],"all":[103],"layers":[104],"or":[105],"only":[106],"final":[108],"layer.Additionally,":[109],"present":[111],"implicit":[113],"contrastive":[115],"during":[117],"inference":[118],"away":[122],"toxic":[124],"outputs.Extensive":[125],"experiments":[126],"natural":[128],"language":[129],"understanding,":[130],"reasoning,":[131],"generation":[133],"tasks,":[134],"conducted":[135],"GPT-2,":[137],"Llama2-7B,":[138],"Mistral-7B,":[140],"demonstrate":[141],"effectiveness":[143],"efficiency":[145,160],"our":[147],"approach.As":[148],"model-agnostic":[150],"framework,":[151],"it":[152],"enhances":[153],"while":[158],"improving":[159],"safe":[162],"deployment.":[163]},"counts_by_year":[],"updated_date":"2026-07-14T23:27:15.235271","created_date":"2025-10-10T00:00:00"}