{"id":"https://openalex.org/W7160516884","doi":"https://doi.org/10.48550/arxiv.2605.04072","title":"Sparse Autoencoder Decomposition of Clinical Sequence Model Representations: Feature Complexity, Task Specialisation, and Mortality Prediction","display_name":"Sparse Autoencoder Decomposition of Clinical Sequence Model Representations: Feature Complexity, Task Specialisation, and Mortality Prediction","publication_year":2026,"publication_date":"2026-04-13","ids":{"openalex":"https://openalex.org/W7160516884","doi":"https://doi.org/10.48550/arxiv.2605.04072"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.04072","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.04072","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":null,"license_id":null,"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.2605.04072","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5112250373","display_name":"Chris Sainsbury","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Sainsbury, Chris","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135568785","display_name":"Feng Dong","orcid":"https://orcid.org/0009-0005-2941-5706"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Dong, Feng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5026210150","display_name":"Andreas Karwath","orcid":"https://orcid.org/0000-0002-6942-3760"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Karwath, Andreas","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/T13702","display_name":"Machine Learning in Healthcare","score":0.9272000193595886,"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/T13702","display_name":"Machine Learning in Healthcare","score":0.9272000193595886,"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/T11636","display_name":"Artificial Intelligence in Healthcare and Education","score":0.01769999973475933,"subfield":{"id":"https://openalex.org/subfields/2718","display_name":"Health Informatics"},"field":{"id":"https://openalex.org/fields/27","display_name":"Medicine"},"domain":{"id":"https://openalex.org/domains/4","display_name":"Health Sciences"}},{"id":"https://openalex.org/T11642","display_name":"Genomics and Rare Diseases","score":0.016200000420212746,"subfield":{"id":"https://openalex.org/subfields/1311","display_name":"Genetics"},"field":{"id":"https://openalex.org/fields/13","display_name":"Biochemistry, Genetics and Molecular Biology"},"domain":{"id":"https://openalex.org/domains/1","display_name":"Life Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/autoencoder","display_name":"Autoencoder","score":0.5830000042915344},{"id":"https://openalex.org/keywords/residual","display_name":"Residual","score":0.5486999750137329},{"id":"https://openalex.org/keywords/autoregressive-model","display_name":"Autoregressive model","score":0.5353999733924866},{"id":"https://openalex.org/keywords/transformer","display_name":"Transformer","score":0.5273000001907349},{"id":"https://openalex.org/keywords/pattern-recognition","display_name":"Pattern recognition (psychology)","score":0.4690999984741211},{"id":"https://openalex.org/keywords/security-token","display_name":"Security token","score":0.44359999895095825},{"id":"https://openalex.org/keywords/false-positive-paradox","display_name":"False positive paradox","score":0.38429999351501465},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.357699990272522}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.645799994468689},{"id":"https://openalex.org/C101738243","wikidata":"https://www.wikidata.org/wiki/Q786435","display_name":"Autoencoder","level":3,"score":0.5830000042915344},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5684999823570251},{"id":"https://openalex.org/C155512373","wikidata":"https://www.wikidata.org/wiki/Q287450","display_name":"Residual","level":2,"score":0.5486999750137329},{"id":"https://openalex.org/C159877910","wikidata":"https://www.wikidata.org/wiki/Q2202883","display_name":"Autoregressive model","level":2,"score":0.5353999733924866},{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.5273000001907349},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.4690999984741211},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.44359999895095825},{"id":"https://openalex.org/C64869954","wikidata":"https://www.wikidata.org/wiki/Q1859747","display_name":"False positive paradox","level":2,"score":0.38429999351501465},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.357699990272522},{"id":"https://openalex.org/C52622490","wikidata":"https://www.wikidata.org/wiki/Q1026626","display_name":"Feature extraction","level":2,"score":0.35420000553131104},{"id":"https://openalex.org/C2776401178","wikidata":"https://www.wikidata.org/wiki/Q12050496","display_name":"Feature (linguistics)","level":2,"score":0.3522999882698059},{"id":"https://openalex.org/C2778112365","wikidata":"https://www.wikidata.org/wiki/Q3511065","display_name":"Sequence (biology)","level":2,"score":0.35109999775886536},{"id":"https://openalex.org/C64341305","wikidata":"https://www.wikidata.org/wiki/Q4919225","display_name":"Bivariate analysis","level":2,"score":0.32989999651908875},{"id":"https://openalex.org/C2780186347","wikidata":"https://www.wikidata.org/wiki/Q11414","display_name":"Subnetwork","level":2,"score":0.31850001215934753},{"id":"https://openalex.org/C148524875","wikidata":"https://www.wikidata.org/wiki/Q6975395","display_name":"F1 score","level":2,"score":0.3075999915599823},{"id":"https://openalex.org/C124304363","wikidata":"https://www.wikidata.org/wiki/Q673661","display_name":"Abstraction","level":2,"score":0.3005000054836273},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.29989999532699585},{"id":"https://openalex.org/C28490314","wikidata":"https://www.wikidata.org/wiki/Q189436","display_name":"Speech recognition","level":1,"score":0.29490000009536743},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.2646999955177307},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.2644999921321869}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.04072","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.04072","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2605.04072","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.04072","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"sustainable_development_goals":[{"score":0.6585440635681152,"display_name":"Good health and well-being","id":"https://metadata.un.org/sdg/3"}],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Sparse":[0],"autoencoders":[1],"(SAEs)":[2],"have":[3],"been":[4],"applied":[5],"to":[6,17,114],"large":[7],"language":[8,12],"models":[9],"and":[10,46,197],"protein":[11],"models,":[13],"but":[14,186],"not":[15,112,187],"systematically":[16],"electronic":[18],"health":[19],"record":[20],"(EHR)":[21],"foundation":[22],"models.":[23],"We":[24],"train":[25],"TopK":[26],"SAEs":[27],"on":[28,43,145],"FlatASCEND,":[29],"a":[30,106],"14.5-million-parameter":[31],"autoregressive":[32],"clinical":[33,75],"sequence":[34],"model,":[35],"at":[36],"all":[37,128],"10":[38],"residual":[39],"stream":[40],"extraction":[41],"points":[42],"INSPECT":[44,152],"(outpatient)":[45],"MIMIC-IV":[47],"(ICU).":[48],"SAE":[49,84,125,134,142,162],"decomposition":[50],"reveals":[51],"progressive":[52],"abstraction":[53],"across":[54,73,127,192],"transformer":[55],"depth:":[56],"layer-0":[57],"features":[58,67,85,126,199],"are":[59,176],"near-perfect":[60],"token":[61,71],"detectors":[62],"(45.7%":[63],"singleton),":[64],"while":[65,94],"layer-6":[66],"span":[68],"approximately":[69],"30":[70],"types":[72],"multiple":[74],"categories":[76],"(0.5%":[77],"singleton).":[78],"Under":[79],"full-sequence":[80],"simple":[81],"linear":[82],"probes,":[83],"outperform":[86,97],"dense":[87,95,120,137],"representations":[88,96,121],"for":[89,98],"discrete":[90],"event":[91],"prediction":[92,101],"(mortality)":[93],"continuous":[99],"magnitude":[100],"(length":[102],"of":[103,183],"stay)":[104],"-":[105],"probe-level":[107],"representational":[108],"phenomenon":[109],"that":[110],"does":[111],"extend":[113],"clinically":[115],"relevant":[116],"leakage-safe":[117],"windows,":[118],"where":[119],"match":[122],"or":[123],"exceed":[124],"tested":[129],"settings":[130],"(eICU-CRD":[131],"48-hour":[132],"AUC:":[133],"0.871":[135],"versus":[136,150,155],"0.880;":[138],"base":[139],"model":[140],"zero-shot,":[141],"dictionaries":[143],"trained":[144],"eICU":[146],"activations;":[147],"MIMIC-IV:":[148],"0.836":[149],"0.914;":[151],"1-year/3-year:":[153],"0.697":[154],"0.800).":[156],"A":[157],"delta-mode":[158],"intervention":[159],"method":[160],"reduces":[161],"perturbation":[163,174],"noise":[164],"by":[165],"86x,":[166],"enabling":[167],"cleaner":[168],"feature-level":[169],"experiments,":[170],"though":[171],"the":[172],"resulting":[173],"effects":[175],"larger":[177],"than":[178,206],"random":[179,193],"controls":[180],"in":[181],"3":[182],"4":[184],"conditions":[185],"formally":[188],"significant.":[189],"Feature":[190],"reproducibility":[191],"seeds":[194],"is":[195],"21%,":[196],"individual":[198],"should":[200],"be":[201],"interpreted":[202],"as":[203],"illustrative":[204],"rather":[205],"stable.":[207]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-05-08T00:00:00"}