{"id":"https://openalex.org/W4415320092","doi":"https://doi.org/10.48550/arxiv.2504.18274","title":"Structural Inference: Interpreting Small Language Models with Susceptibilities","display_name":"Structural Inference: Interpreting Small Language Models with Susceptibilities","alternate_titles":[],"publication_year":2025,"publication_date":"2025-04-25","ids":{"openalex":"https://openalex.org/W4415320092","doi":"https://doi.org/10.48550/arxiv.2504.18274"},"language":"en","primary_location":{"id":"pmh:oai:proceedings.iclr.cc:2026/57eb852970a17cf9c23d35b48fbfd191","doi":null,"is_oa":true,"landing_page_url":"https://proceedings.iclr.cc/paper_files/paper/2026/hash/57eb852970a17cf9c23d35b48fbfd191-Abstract-Conference.html","pdf_url":null,"source":{"id":"https://openalex.org/S4306419637","display_name":"International Conference on Learning Representations","issn_l":null,"issn":null,"is_oa":true,"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":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"International Conference on Learning Representations 2026","raw_type":"info:eu-repo/semantics/conferenceObject"},"parent_works":[],"lineage":["https://openalex.org/W4415320092"],"child_works_count":0,"type":"conference-paper","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://proceedings.iclr.cc/paper_files/paper/2026/hash/57eb852970a17cf9c23d35b48fbfd191-Abstract-Conference.html","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":null,"display_name":"Baker, Garrett","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Baker, Garrett","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Wang, George","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, George","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5104240187","display_name":"Jesse Hoogland","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Hoogland, Jesse","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5082251796","display_name":"Daniel Murfet","orcid":"https://orcid.org/0000-0003-2495-0070","observed_orcids":["https://orcid.org/0000-0003-2495-0070"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Murfet, Daniel","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":2.9019,"has_fulltext":true,"cited_by_count":15,"citation_normalized_percentile":{"value":0.92219848,"is_in_top_1_percent":false,"is_in_top_10_percent":true},"cited_by_percentile_year":{"min":97,"max":99},"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_pulled":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":1.0,"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/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":1.0,"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/T10028","display_name":"Topic Modeling","score":0.0,"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":0.0,"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/small-language-models","display_name":"small language models","score":0.7670000195503235},{"id":"https://openalex.org/keywords/interpretability","display_name":"interpretability","score":0.9319999814033508},{"id":"https://openalex.org/keywords/linear-response","display_name":"linear response","score":0.6669999957084656},{"id":"https://openalex.org/keywords/susceptibility","display_name":"susceptibility","score":0.8080000281333923},{"id":"https://openalex.org/keywords/neural-network-interpretability","display_name":"neural network interpretability","score":0.5509999990463257},{"id":"https://openalex.org/keywords/response-matrix","display_name":"response matrix","score":0.699999988079071},{"id":"https://openalex.org/keywords/low-rank-structure","display_name":"low-rank structure","score":0.9300000071525574},{"id":"https://openalex.org/keywords/functional-modules","display_name":"functional modules","score":0.5609999895095825},{"id":"https://openalex.org/keywords/induction-heads","display_name":"induction heads","score":0.8999999761581421}],"concepts":[{"id":"https://openalex.org/C2781067378","wikidata":"https://www.wikidata.org/wiki/Q17027399","display_name":"Interpretability","level":2,"score":0.9229000210762024},{"id":"https://openalex.org/C107673813","wikidata":"https://www.wikidata.org/wiki/Q812534","display_name":"Bayesian probability","level":2,"score":0.5601000189781189},{"id":"https://openalex.org/C32848918","wikidata":"https://www.wikidata.org/wiki/Q845789","display_name":"Observable","level":2,"score":0.5195000171661377},{"id":"https://openalex.org/C168167062","wikidata":"https://www.wikidata.org/wiki/Q1117970","display_name":"Component (thermodynamics)","level":2,"score":0.5171999931335449},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.47279998660087585},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.45680001378059387},{"id":"https://openalex.org/C114289077","wikidata":"https://www.wikidata.org/wiki/Q3284399","display_name":"Statistical model","level":2,"score":0.4253999888896942},{"id":"https://openalex.org/C163175372","wikidata":"https://www.wikidata.org/wiki/Q3339222","display_name":"Linear model","level":2,"score":0.40220001339912415},{"id":"https://openalex.org/C177918212","wikidata":"https://www.wikidata.org/wiki/Q803623","display_name":"Perturbation (astronomy)","level":2,"score":0.36820000410079956},{"id":"https://openalex.org/C2986587452","wikidata":"https://www.wikidata.org/wiki/Q938438","display_name":"Statistical analysis","level":2,"score":0.3422999978065491},{"id":"https://openalex.org/C106487976","wikidata":"https://www.wikidata.org/wiki/Q685816","display_name":"Matrix (chemical analysis)","level":2,"score":0.34049999713897705},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.3188999891281128},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.3149999976158142},{"id":"https://openalex.org/C33724603","wikidata":"https://www.wikidata.org/wiki/Q812540","display_name":"Bayesian network","level":2,"score":0.3034999966621399},{"id":"https://openalex.org/C50644808","wikidata":"https://www.wikidata.org/wiki/Q192776","display_name":"Artificial neural network","level":2,"score":0.30059999227523804},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.28439998626708984}],"mesh":[],"locations_count":3,"locations":[{"id":"pmh:oai:proceedings.iclr.cc:2026/57eb852970a17cf9c23d35b48fbfd191","doi":null,"is_oa":true,"landing_page_url":"https://proceedings.iclr.cc/paper_files/paper/2026/hash/57eb852970a17cf9c23d35b48fbfd191-Abstract-Conference.html","pdf_url":null,"source":{"id":"https://openalex.org/S4306419637","display_name":"International Conference on Learning Representations","issn_l":null,"issn":null,"is_oa":true,"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":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"International Conference on Learning Representations 2026","raw_type":"info:eu-repo/semantics/conferenceObject"},{"id":"pmh:oai:arXiv.org:2504.18274","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2504.18274","pdf_url":"https://arxiv.org/pdf/2504.18274","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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2504.18274","doi":"https://doi.org/10.48550/arxiv.2504.18274","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2504.18274","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":"pmh:oai:proceedings.iclr.cc:2026/57eb852970a17cf9c23d35b48fbfd191","doi":null,"is_oa":true,"landing_page_url":"https://proceedings.iclr.cc/paper_files/paper/2026/hash/57eb852970a17cf9c23d35b48fbfd191-Abstract-Conference.html","pdf_url":null,"source":{"id":"https://openalex.org/S4306419637","display_name":"International Conference on Learning Representations","issn_l":null,"issn":null,"is_oa":true,"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":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"International Conference on Learning Representations 2026","raw_type":"info:eu-repo/semantics/conferenceObject"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[{"id":"https://openalex.org/F4320315389","display_name":"Open Philanthropy Project","ror":"https://ror.org/004d1k391"}],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4415320092.pdf","grobid_xml":"https://content.openalex.org/works/W4415320092.grobid-xml"},"referenced_works_count":43,"referenced_works":["https://openalex.org/W1538609674","https://openalex.org/W1553535237","https://openalex.org/W1992129502","https://openalex.org/W2054316680","https://openalex.org/W2124641450","https://openalex.org/W2126377586","https://openalex.org/W2132589721","https://openalex.org/W2151257151","https://openalex.org/W2167433878","https://openalex.org/W2243651912","https://openalex.org/W2292751606","https://openalex.org/W2621461287","https://openalex.org/W2785648721","https://openalex.org/W2806956262","https://openalex.org/W2882319491","https://openalex.org/W2898365215","https://openalex.org/W2963410064","https://openalex.org/W3014794741","https://openalex.org/W3118781290","https://openalex.org/W4249882522","https://openalex.org/W4253667136","https://openalex.org/W4286224767","https://openalex.org/W4297412003","https://openalex.org/W4308023630","https://openalex.org/W4316135772","https://openalex.org/W4320086400","https://openalex.org/W4376311932","https://openalex.org/W4376652984","https://openalex.org/W4378976798","https://openalex.org/W4385473442","https://openalex.org/W4385572928","https://openalex.org/W4390872014","https://openalex.org/W4391159099","https://openalex.org/W4392182427","https://openalex.org/W4400142056","https://openalex.org/W4403885496","https://openalex.org/W4404783649","https://openalex.org/W4415261756","https://openalex.org/W4415797858","https://openalex.org/W4416379343","https://openalex.org/W6910700162","https://openalex.org/W7133245087","https://openalex.org/W7165557573"],"related_works":["https://openalex.org/W7125123251","https://openalex.org/W7214106579","https://openalex.org/W7162314614","https://openalex.org/W4411638659","https://openalex.org/W7160844668","https://openalex.org/W4406122373","https://openalex.org/W6891855200","https://openalex.org/W7125680013","https://openalex.org/W7127084415","https://openalex.org/W7204934723"],"abstract_inverted_index":{"We":[0,77],"develop":[1],"a":[2,10,14,37,49,82,98],"linear":[3],"response":[4,83],"framework":[5],"for":[6,26],"interpretability":[7],"that":[8,72],"treats":[9],"neural":[11],"network":[12],"as":[13,74,92],"Bayesian":[15],"statistical":[16],"mechanical":[17],"system.":[18],"A":[19],"small":[20],"perturbation":[21],"of":[22,44,52],"the":[23,29,41,53],"data":[24],"distribution,":[25],"example":[27],"shifting":[28],"Pile":[30],"toward":[31],"GitHub":[32],"or":[33],"legal":[34],"text,":[35],"induces":[36],"first-order":[38],"change":[39],"in":[40,97],"posterior":[42],"expectation":[43],"an":[45],"observable":[46],"localized":[47],"on":[48],"chosen":[50],"component":[51],"network.":[54],"The":[55],"resulting":[56],"susceptibility":[57],"can":[58],"be":[59],"estimated":[60],"efficiently":[61],"with":[62],"local":[63],"SGLD":[64],"samples":[65],"and":[66,94],"factorizes":[67],"into":[68,81],"signed,":[69],"per-token":[70],"contributions":[71],"serve":[73],"attribution":[75],"scores.":[76],"combine":[78],"these":[79],"susceptibilities":[80],"matrix":[84],"whose":[85],"low-rank":[86],"structure":[87],"separates":[88],"functional":[89],"modules":[90],"such":[91],"multigram":[93],"induction":[95],"heads":[96],"3M-parameter":[99],"transformer.":[100]},"cited_by_api_url":"https://api.openalex.org/works?filter=cites:W4415320092","counts_by_year":[{"year":2026,"cited_by_count":10},{"year":2025,"cited_by_count":5}],"updated_date":"2026-11-19T14:11:59.513116","created_date":"2025-10-18T00:00:00"}
