{"id":"https://openalex.org/W7140141392","doi":"https://doi.org/10.18653/v1/2026.findings-eacl.40","title":"Denoising Concept Vectors with Sparse Autoencoders for Improved Language Model Steering","display_name":"Denoising Concept Vectors with Sparse Autoencoders for Improved Language Model Steering","publication_year":2026,"publication_date":"2026-01-01","ids":{"openalex":"https://openalex.org/W7140141392","doi":"https://doi.org/10.18653/v1/2026.findings-eacl.40"},"language":"en","primary_location":{"id":"doi:10.18653/v1/2026.findings-eacl.40","doi":"https://doi.org/10.18653/v1/2026.findings-eacl.40","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-eacl.40","pdf_url":"https://aclanthology.org/2026.findings-eacl.40.pdf","source":{"id":"https://openalex.org/S7407087051","display_name":"Findings of the Association for Computational Linguistics: EACL","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"conference"},"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: EACL 2026","raw_type":"proceedings-article"},"volume":null,"type":"conference-paper","indexed_in":["arxiv","crossref","datacite"],"open_access":{"is_oa":true,"oa_status":"hybrid","oa_url":"https://aclanthology.org/2026.findings-eacl.40.pdf","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5113144161","display_name":"Haiyan Zhao","orcid":"https://orcid.org/0009-0006-5358-6895","observed_orcids":["https://orcid.org/0009-0006-5358-6895"]},"institutions":[{"id":"https://openalex.org/I118118575","display_name":"New Jersey Institute of Technology","ror":"https://ror.org/05e74xb87","country_code":"US","type":"education","lineage":["https://openalex.org/I118118575"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Haiyan Zhao","raw_affiliation_strings":["New Jersey Institute of Technology"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"New Jersey Institute of Technology","institution_ids":["https://openalex.org/I118118575"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5089884284","display_name":"Xuansheng Wu","orcid":"https://orcid.org/0000-0002-7816-7658","observed_orcids":["https://orcid.org/0000-0002-7816-7658"]},"institutions":[{"id":"https://openalex.org/I165733156","display_name":"University of Georgia","ror":"https://ror.org/00te3t702","country_code":"US","type":"education","lineage":["https://openalex.org/I165733156"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Xuansheng Wu","raw_affiliation_strings":["University of Georgia"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Georgia","institution_ids":["https://openalex.org/I165733156"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5130355008","display_name":"Fan Yang","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I47251452","display_name":"Wake Forest University","ror":"https://ror.org/0207ad724","country_code":"US","type":"education","lineage":["https://openalex.org/I47251452"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Fan Yang","raw_affiliation_strings":["Wake Forest University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Wake Forest University","institution_ids":["https://openalex.org/I47251452"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101364672","display_name":"Bo Shen","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I118118575","display_name":"New Jersey Institute of Technology","ror":"https://ror.org/05e74xb87","country_code":"US","type":"education","lineage":["https://openalex.org/I118118575"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Bo Shen","raw_affiliation_strings":["New Jersey Institute of Technology"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"New Jersey Institute of Technology","institution_ids":["https://openalex.org/I118118575"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5007489034","display_name":"Ninghao Liu","orcid":"https://orcid.org/0000-0002-9170-2424","observed_orcids":["https://orcid.org/0000-0002-9170-2424","https://orcid.org/0009-0002-8737-420X"]},"institutions":[{"id":"https://openalex.org/I165733156","display_name":"University of Georgia","ror":"https://ror.org/00te3t702","country_code":"US","type":"education","lineage":["https://openalex.org/I165733156"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Ninghao Liu","raw_affiliation_strings":["University of Georgia"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Georgia","institution_ids":["https://openalex.org/I165733156"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5129979709","display_name":"Mengnan Du","orcid":"https://orcid.org/0000-0002-1614-6069","observed_orcids":["https://orcid.org/0000-0002-1614-6069"]},"institutions":[{"id":"https://openalex.org/I118118575","display_name":"New Jersey Institute of Technology","ror":"https://ror.org/05e74xb87","country_code":"US","type":"education","lineage":["https://openalex.org/I118118575"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Mengnan Du","raw_affiliation_strings":["New Jersey Institute of Technology"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"New Jersey Institute of Technology","institution_ids":["https://openalex.org/I118118575"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":3,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":{"value":0.15622109,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"797","last_page":"808"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.9038000106811523,"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":0.9038000106811523,"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":0.06159999966621399,"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/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.010599999688565731,"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/concept-vectors","display_name":"concept vectors","score":0.6740000247955322},{"id":"https://openalex.org/keywords/sparse-autoencoders","display_name":"sparse autoencoders","score":0.9409999847412109},{"id":"https://openalex.org/keywords/noise-reduction","display_name":"noise reduction","score":0.6549999713897705},{"id":"https://openalex.org/keywords/linear-probing","display_name":"linear probing","score":0.7559999823570251},{"id":"https://openalex.org/keywords/large-language-models","display_name":"large language models","score":0.652999997138977}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6503999829292297},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.6294000148773193},{"id":"https://openalex.org/C163294075","wikidata":"https://www.wikidata.org/wiki/Q581861","display_name":"Noise reduction","level":2,"score":0.4896000027656555},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.39340001344680786},{"id":"https://openalex.org/C99498987","wikidata":"https://www.wikidata.org/wiki/Q2210247","display_name":"Noise (video)","level":3,"score":0.37220001220703125},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.3671000003814697},{"id":"https://openalex.org/C31972630","wikidata":"https://www.wikidata.org/wiki/Q844240","display_name":"Computer vision","level":1,"score":0.34439998865127563},{"id":"https://openalex.org/C2776401178","wikidata":"https://www.wikidata.org/wiki/Q12050496","display_name":"Feature (linguistics)","level":2,"score":0.3154999911785126},{"id":"https://openalex.org/C111335779","wikidata":"https://www.wikidata.org/wiki/Q3454686","display_name":"Reduction (mathematics)","level":2,"score":0.2989000082015991},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.26669999957084656},{"id":"https://openalex.org/C67186912","wikidata":"https://www.wikidata.org/wiki/Q367664","display_name":"Data modeling","level":2,"score":0.26409998536109924},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.25870001316070557},{"id":"https://openalex.org/C28490314","wikidata":"https://www.wikidata.org/wiki/Q189436","display_name":"Speech recognition","level":1,"score":0.25769999623298645}],"mesh":[],"locations_count":3,"locations":[{"id":"doi:10.18653/v1/2026.findings-eacl.40","doi":"https://doi.org/10.18653/v1/2026.findings-eacl.40","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-eacl.40","pdf_url":"https://aclanthology.org/2026.findings-eacl.40.pdf","source":{"id":"https://openalex.org/S7407087051","display_name":"Findings of the Association for Computational Linguistics: EACL","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"conference"},"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: EACL 2026","raw_type":"proceedings-article"},{"id":"pmh:oai:arXiv.org:2505.15038","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2505.15038","pdf_url":"https://arxiv.org/pdf/2505.15038","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":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2505.15038","doi":"https://doi.org/10.48550/arxiv.2505.15038","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2505.15038","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":"doi:10.18653/v1/2026.findings-eacl.40","doi":"https://doi.org/10.18653/v1/2026.findings-eacl.40","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-eacl.40","pdf_url":"https://aclanthology.org/2026.findings-eacl.40.pdf","source":{"id":"https://openalex.org/S7407087051","display_name":"Findings of the Association for Computational Linguistics: EACL","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"conference"},"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: EACL 2026","raw_type":"proceedings-article"},"sustainable_development_goals":[],"sustainable_development_goals_aurora":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7140141392.pdf","grobid_xml":"https://content.openalex.org/works/W7140141392.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4416366301","https://openalex.org/W4416037088","https://openalex.org/W4417296956","https://openalex.org/W4406961517","https://openalex.org/W7221747770","https://openalex.org/W4415112074","https://openalex.org/W7170057469","https://openalex.org/W7127326285","https://openalex.org/W4404354935","https://openalex.org/W7166821494"],"abstract_inverted_index":{"Linear":[0],"concept":[1],"vectors":[2],"effectively":[3],"steer":[4],"LLMs,":[5],"but":[6],"existing":[7],"methods":[8],"suffer":[9],"from":[10,48],"noisy":[11],"features":[12],"in":[13],"diverse":[14],"datasets":[15],"that":[16,41,58],"undermine":[17],"steering":[18,73],"robustness.We":[19],"propose":[20],"Sparse":[21],"Autoencoder-Denoised":[22],"Concept":[23],"Vectors":[24],"(SDCV),":[25],"which":[26],"selectively":[27],"keep":[28],"the":[29],"most":[30],"discriminative":[31],"SAE":[32],"latents":[33,57],"while":[34,82],"reconstructing":[35],"hidden":[36],"representations.Our":[37],"key":[38],"insight":[39],"is":[40],"concept-relevant":[42],"signals":[43],"can":[44],"be":[45],"explicitly":[46],"separated":[47],"dataset":[49],"noise":[50],"by":[51,76],"scaling":[52],"up":[53],"activations":[54],"of":[55],"top-k":[56],"best":[59],"differentiate":[60],"positive":[61],"and":[62,68],"negative":[63],"samples.Applied":[64],"to":[65],"linear":[66],"probing":[67],"differencein-mean,":[69],"SDCV":[70],"consistently":[71],"improves":[72],"success":[74],"rates":[75],"4-16%":[77],"across":[78],"six":[79],"challenging":[80],"concepts,":[81],"maintaining":[83],"topic":[84],"relevance.":[85]},"counts_by_year":[],"updated_date":"2026-10-16T09:59:22.235173","created_date":"2026-03-24T00:00:00"}