{"id":"https://openalex.org/W4412945148","doi":"https://doi.org/10.18653/v1/2025.acl-long.1209","title":"Math Neurosurgery: Isolating Language Models\u2019 Math Reasoning Abilities Using Only Forward Passes","display_name":"Math Neurosurgery: Isolating Language Models\u2019 Math Reasoning Abilities Using Only Forward Passes","publication_year":2025,"publication_date":"2025-01-01","ids":{"openalex":"https://openalex.org/W4412945148","doi":"https://doi.org/10.18653/v1/2025.acl-long.1209"},"language":"en","primary_location":{"id":"doi:10.18653/v1/2025.acl-long.1209","doi":"https://doi.org/10.18653/v1/2025.acl-long.1209","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.acl-long.1209","pdf_url":"https://aclanthology.org/2025.acl-long.1209.pdf","source":{"id":"https://openalex.org/S4393915740","display_name":"Proceedings of the conference - Association for Computational Linguistics. Meeting","issn_l":"0736-587X","issn":["0736-587X"],"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":["norway-1"],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"conference"},"license":"other-oa","license_id":"https://openalex.org/licenses/other-oa","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","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/2025.acl-long.1209.pdf","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5093242240","display_name":"Bryan R. Christ","orcid":"https://orcid.org/0009-0000-4356-7961","observed_orcids":["https://orcid.org/0009-0000-4356-7961"]},"institutions":[{"id":"https://openalex.org/I51556381","display_name":"University of Virginia","ror":"https://ror.org/0153tk833","country_code":"US","type":"education","lineage":["https://openalex.org/I51556381"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Bryan R Christ","raw_affiliation_strings":["University of Virginia"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Virginia","institution_ids":["https://openalex.org/I51556381"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5140833198","display_name":"Zachary Gottesman","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I51556381","display_name":"University of Virginia","ror":"https://ror.org/0153tk833","country_code":"US","type":"education","lineage":["https://openalex.org/I51556381"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Zachary Gottesman","raw_affiliation_strings":["University of Virginia"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Virginia","institution_ids":["https://openalex.org/I51556381"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5015466498","display_name":"Jonathan Kropko","orcid":"https://orcid.org/0000-0003-3403-7648","observed_orcids":["https://orcid.org/0000-0003-3403-7648"]},"institutions":[{"id":"https://openalex.org/I51556381","display_name":"University of Virginia","ror":"https://ror.org/0153tk833","country_code":"US","type":"education","lineage":["https://openalex.org/I51556381"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Jonathan Kropko","raw_affiliation_strings":["University of Virginia"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Virginia","institution_ids":["https://openalex.org/I51556381"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5075881948","display_name":"Thomas Hartvigsen","orcid":"https://orcid.org/0000-0002-5288-2792","observed_orcids":["https://orcid.org/0000-0002-5288-2792"]},"institutions":[{"id":"https://openalex.org/I51556381","display_name":"University of Virginia","ror":"https://ror.org/0153tk833","country_code":"US","type":"education","lineage":["https://openalex.org/I51556381"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Thomas Hartvigsen","raw_affiliation_strings":["University of Virginia"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Virginia","institution_ids":["https://openalex.org/I51556381"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I51556381"],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":{"value":0.1601049,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"24803","last_page":"24840"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.3458999991416931,"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":0.3458999991416931,"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":0.0820000022649765,"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":0.0803999975323677,"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/language-models","display_name":"language models","score":0.6299999952316284},{"id":"https://openalex.org/keywords/mathematical-reasoning","display_name":"mathematical reasoning","score":0.9350000023841858},{"id":"https://openalex.org/keywords/forward-pass","display_name":"forward pass","score":0.546999990940094},{"id":"https://openalex.org/keywords/large-language-models","display_name":"large language models","score":0.4490000009536743}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.4781716465950012},{"id":"https://openalex.org/C145420912","wikidata":"https://www.wikidata.org/wiki/Q853077","display_name":"Mathematics education","level":1,"score":0.4393078088760376},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.3022542893886566}],"mesh":[],"locations_count":3,"locations":[{"id":"doi:10.18653/v1/2025.acl-long.1209","doi":"https://doi.org/10.18653/v1/2025.acl-long.1209","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.acl-long.1209","pdf_url":"https://aclanthology.org/2025.acl-long.1209.pdf","source":{"id":"https://openalex.org/S4393915740","display_name":"Proceedings of the conference - Association for Computational Linguistics. Meeting","issn_l":"0736-587X","issn":["0736-587X"],"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":["norway-1"],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"conference"},"license":"other-oa","license_id":"https://openalex.org/licenses/other-oa","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","raw_type":"proceedings-article"},{"id":"pmh:oai:arXiv.org:2410.16930","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2410.16930","pdf_url":"https://arxiv.org/pdf/2410.16930","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-sa","license_id":"https://openalex.org/licenses/cc-by-sa","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2410.16930","doi":"https://doi.org/10.48550/arxiv.2410.16930","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2410.16930","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.18653/v1/2025.acl-long.1209","doi":"https://doi.org/10.18653/v1/2025.acl-long.1209","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.acl-long.1209","pdf_url":"https://aclanthology.org/2025.acl-long.1209.pdf","source":{"id":"https://openalex.org/S4393915740","display_name":"Proceedings of the conference - Association for Computational Linguistics. Meeting","issn_l":"0736-587X","issn":["0736-587X"],"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":["norway-1"],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"conference"},"license":"other-oa","license_id":"https://openalex.org/licenses/other-oa","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","raw_type":"proceedings-article"},"sustainable_development_goals":[],"sustainable_development_goals_aurora":[{"display_name":"Quality education","id":"https://metadata.un.org/sdg/4","score":0.6100000143051147}],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4412945148.pdf","grobid_xml":"https://content.openalex.org/works/W4412945148.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4402670720","https://openalex.org/W4399115472","https://openalex.org/W6929401685","https://openalex.org/W7092284188","https://openalex.org/W4386044080","https://openalex.org/W7165129305","https://openalex.org/W7095006324","https://openalex.org/W4400341393","https://openalex.org/W7221035608","https://openalex.org/W4386554674"],"abstract_inverted_index":{"Math":[0,76],"reasoning":[1,35,130],"is":[2,14,36,44,166],"an":[3],"active":[4],"area":[5],"of":[6,17,69,171],"Large":[7],"Language":[8],"Model":[9],"(LLM)":[10],"research":[11],"because":[12],"it":[13,43],"a":[15,45,79,127,143,147,180],"hallmark":[16],"artificial":[18],"intelligence":[19],"and":[20,41,66,102,157],"has":[21],"implications":[22],"in":[23,89],"several":[24],"domains,":[25],"including":[26],"math":[27,34,60,73,129],"education.However,":[28],"few":[29],"works":[30],"have":[31],"explored":[32],"how":[33,70],"encoded":[37],"within":[38,51],"LLM":[39],"parameters":[40,88,111,122,141,178],"if":[42],"skill":[46],"that":[47],"can":[48],"be":[49],"isolated":[50],"models.Doing":[52],"so":[53],"could":[54],"allow":[55],"targeted":[56],"intervention":[57],"to":[58,85,104,189],"improve":[59],"performance":[61,152],"without":[62,132],"altering":[63],"non-math":[64,163],"behavior":[65,164],"foster":[67],"understanding":[68],"models":[71],"encode":[72],"reasoning.We":[74],"introduce":[75],"Neurosurgery":[77],"(MathNeuro),":[78],"computationally":[80],"efficient":[81],"method":[82],"we":[83,125],"use":[84],"isolate":[86],"math-specific":[87,110,177,192],"LLMs":[90],"using":[91,100,179],"only":[92],"forward":[93],"passes.MathNeuro":[94],"builds":[95],"on":[96,155,159,191],"existing":[97],"work":[98,188],"by":[99,112,142,153],"weights":[101],"activations":[103],"calculate":[105],"parameter":[106],"importance,":[107],"but":[108],"isolates":[109],"filtering":[113],"out":[114],"those":[115],"important":[116],"for":[117,186],"general":[118,136],"language":[119,137],"tasks.Through":[120],"pruning":[121],"MathNeuro":[123],"identifies,":[124],"delete":[126],"LLM's":[128,151],"ability":[131],"significantly":[133],"impacting":[134],"its":[135,172],"ability.Scaling":[138],"the":[139,184],"identified":[140],"small":[144],"constant":[145],"improves":[146],"pretrained":[148],"or":[149],"instruction-tuned":[150],"4-17%":[154],"GSM8K":[156],"5-35%":[158],"MATH":[160],"while":[161],"leaving":[162],"unaltered.Math-Neuro":[165],"also":[167],"data":[168],"efficient:":[169],"most":[170],"effectiveness":[173],"holds":[174],"when":[175],"identifying":[176],"single":[181],"sample.MathNeuro":[182],"highlights":[183],"potential":[185],"future":[187],"intervene":[190],"parameters.":[193],"1":[194]},"counts_by_year":[],"updated_date":"2026-10-29T09:56:32.521200","created_date":"2025-10-10T00:00:00"}