{"id":"https://openalex.org/W7163919490","doi":"https://doi.org/10.48550/arxiv.2606.07007","title":"A Geometric View for Understanding Concept Learning and Neuron Interpretation in Sparse Autoencoders","display_name":"A Geometric View for Understanding Concept Learning and Neuron Interpretation in Sparse Autoencoders","publication_year":2026,"publication_date":"2026-06-05","ids":{"openalex":"https://openalex.org/W7163919490","doi":"https://doi.org/10.48550/arxiv.2606.07007"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2606.07007","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.07007","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2606.07007","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5138186927","display_name":"Chenhao Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Chenhao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5138162196","display_name":"Chris Lin","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lin, Chris","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5138182554","display_name":"Su-In Lee","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lee, Su-In","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":false,"cited_by_count":0,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":0.32760000228881836,"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":0.32760000228881836,"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/T10775","display_name":"Generative Adversarial Networks and Image Synthesis","score":0.13349999487400055,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"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/T12536","display_name":"Topological and Geometric Data Analysis","score":0.06549999862909317,"subfield":{"id":"https://openalex.org/subfields/1703","display_name":"Computational Theory and Mathematics"},"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/interpretability","display_name":"Interpretability","score":0.9009000062942505},{"id":"https://openalex.org/keywords/interpretation","display_name":"Interpretation (philosophy)","score":0.7210999727249146},{"id":"https://openalex.org/keywords/feature","display_name":"Feature (linguistics)","score":0.6399000287055969},{"id":"https://openalex.org/keywords/artificial-neural-network","display_name":"Artificial neural network","score":0.4693000018596649},{"id":"https://openalex.org/keywords/concept-learning","display_name":"Concept learning","score":0.4521999955177307},{"id":"https://openalex.org/keywords/feature-learning","display_name":"Feature learning","score":0.39899998903274536},{"id":"https://openalex.org/keywords/pattern-recognition","display_name":"Pattern recognition (psychology)","score":0.3474999964237213}],"concepts":[{"id":"https://openalex.org/C2781067378","wikidata":"https://www.wikidata.org/wiki/Q17027399","display_name":"Interpretability","level":2,"score":0.9009000062942505},{"id":"https://openalex.org/C527412718","wikidata":"https://www.wikidata.org/wiki/Q855395","display_name":"Interpretation (philosophy)","level":2,"score":0.7210999727249146},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.697700023651123},{"id":"https://openalex.org/C2776401178","wikidata":"https://www.wikidata.org/wiki/Q12050496","display_name":"Feature (linguistics)","level":2,"score":0.6399000287055969},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6007999777793884},{"id":"https://openalex.org/C50644808","wikidata":"https://www.wikidata.org/wiki/Q192776","display_name":"Artificial neural network","level":2,"score":0.4693000018596649},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.45239999890327454},{"id":"https://openalex.org/C48164120","wikidata":"https://www.wikidata.org/wiki/Q4491893","display_name":"Concept learning","level":2,"score":0.4521999955177307},{"id":"https://openalex.org/C59404180","wikidata":"https://www.wikidata.org/wiki/Q17013334","display_name":"Feature learning","level":2,"score":0.39899998903274536},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.3474999964237213},{"id":"https://openalex.org/C197352929","wikidata":"https://www.wikidata.org/wiki/Q1074074","display_name":"Inductive bias","level":4,"score":0.34709998965263367},{"id":"https://openalex.org/C2986563244","wikidata":"https://www.wikidata.org/wiki/Q6822310","display_name":"Learning to learn","level":2,"score":0.34060001373291016},{"id":"https://openalex.org/C160920958","wikidata":"https://www.wikidata.org/wiki/Q7662746","display_name":"Synthetic data","level":2,"score":0.3384999930858612},{"id":"https://openalex.org/C108583219","wikidata":"https://www.wikidata.org/wiki/Q197536","display_name":"Deep learning","level":2,"score":0.32760000228881836},{"id":"https://openalex.org/C168167062","wikidata":"https://www.wikidata.org/wiki/Q1117970","display_name":"Component (thermodynamics)","level":2,"score":0.3057999908924103},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.29670000076293945},{"id":"https://openalex.org/C83665646","wikidata":"https://www.wikidata.org/wiki/Q42139305","display_name":"Feature vector","level":2,"score":0.2759999930858612},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.2605000138282776},{"id":"https://openalex.org/C47822265","wikidata":"https://www.wikidata.org/wiki/Q854457","display_name":"Complex system","level":2,"score":0.25220000743865967}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2606.07007","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.07007","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2606.07007","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.07007","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"We":[0,41],"propose":[1],"a":[2,7,32,54,101],"unified":[3],"mathematical":[4],"framework":[5],"for":[6,86,104],"geometric":[8,79],"understanding":[9],"of":[10,24,35,46,69,165],"concept":[11,51,121,128,147,171],"learning":[12,28,52,70,122],"and":[13,37,49,59,74,77,83,115,123,138,155,161,168],"neuron":[14,124],"interpretation":[15,125],"in":[16],"sparse":[17,29],"autoencoders":[18],"(SAEs).":[19],"While":[20],"SAEs":[21,157],"improve":[22],"interpretability":[23],"neural":[25],"networks":[26],"by":[27,92,146],"feature":[30,109,111,113],"representations,":[31],"principled":[33],"definition":[34],"''concept''":[36],"''learning''":[38],"remains":[39],"unclear.":[40],"formalize":[42],"concepts":[43,88],"as":[44,53],"sets":[45],"data":[47,152],"points":[48],"cast":[50],"set-alignment":[55],"problem":[56],"between":[57],"human-defined":[58],"model-induced":[60],"concepts.":[61,117],"This":[62],"formulation":[63],"distinguishes":[64],"three":[65],"increasingly":[66],"strong":[67],"notions":[68],"--":[71,76],"detection,":[72],"separation,":[73],"approximation":[75],"yields":[78],"conditions,":[80],"error":[81],"bounds,":[82],"capacity":[84],"constraints":[85],"when":[87],"can":[89,143],"be":[90,144],"represented":[91],"individual":[93],"neurons":[94],"or":[95],"multi-neuron":[96],"units.":[97],"It":[98],"also":[99],"provides":[100],"set-theoretic":[102],"account":[103],"common":[105],"SAE":[106,166],"phenomena,":[107],"including":[108],"splitting,":[110],"absorption,":[112],"families,":[114],"hierarchical":[116],"Finally,":[118],"we":[119],"connect":[120],"through":[126],"formal":[127],"analysis,":[129],"showing":[130],"that":[131,139],"the":[132,159,163],"two":[133],"directions":[134],"need":[135],"not":[136],"agree":[137],"their":[140],"many-to-many":[141],"structure":[142],"organized":[145],"lattices.":[148],"Experiments":[149],"on":[150,170],"synthetic":[151],"with":[153],"ReLU":[154],"Top-$K$":[156],"illustrate":[158],"theory":[160],"reveal":[162],"effects":[164],"size":[167],"sparsity":[169],"learning.":[172]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-06-09T00:00:00"}