{"id":"https://openalex.org/W7167037070","doi":"https://doi.org/10.48550/arxiv.2607.00089","title":"Representation as a Bottleneck for Mechanistic Interpretability: The Manifestation Unit Protocol","display_name":"Representation as a Bottleneck for Mechanistic Interpretability: The Manifestation Unit Protocol","publication_year":2026,"publication_date":"2026-06-30","ids":{"openalex":"https://openalex.org/W7167037070","doi":"https://doi.org/10.48550/arxiv.2607.00089"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2607.00089","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2607.00089","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,"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.2607.00089","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5052624077","display_name":"Hussein Chouman","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chouman, Hussein","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5039494960","display_name":"Wataru Sasaki","orcid":"https://orcid.org/0000-0003-1126-9569"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Sasaki, Wataru","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139908242","display_name":"Tomokazu Matsui","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Matsui, Tomokazu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5005700675","display_name":"Hirohiko Suwa","orcid":"https://orcid.org/0000-0002-8519-3352"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Suwa, Hirohiko","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5019740585","display_name":"Keiichi Yasumoto","orcid":"https://orcid.org/0000-0003-1579-3237"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yasumoto, Keiichi","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":0.25589999556541443,"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":0.25589999556541443,"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/T12808","display_name":"Ferroelectric and Negative Capacitance Devices","score":0.08259999752044678,"subfield":{"id":"https://openalex.org/subfields/2208","display_name":"Electrical and Electronic Engineering"},"field":{"id":"https://openalex.org/fields/22","display_name":"Engineering"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.05469999834895134,"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/interpretability","display_name":"Interpretability","score":0.9398999810218811},{"id":"https://openalex.org/keywords/schema","display_name":"Schema (genetic algorithms)","score":0.5722000002861023},{"id":"https://openalex.org/keywords/discriminative-model","display_name":"Discriminative model","score":0.48260000348091125},{"id":"https://openalex.org/keywords/bottleneck","display_name":"Bottleneck","score":0.4787999987602234},{"id":"https://openalex.org/keywords/encode","display_name":"ENCODE","score":0.4740000069141388},{"id":"https://openalex.org/keywords/external-data-representation","display_name":"External Data Representation","score":0.3384999930858612},{"id":"https://openalex.org/keywords/protocol-data-unit","display_name":"Protocol data unit","score":0.3325999975204468},{"id":"https://openalex.org/keywords/tuple","display_name":"Tuple","score":0.32829999923706055}],"concepts":[{"id":"https://openalex.org/C2781067378","wikidata":"https://www.wikidata.org/wiki/Q17027399","display_name":"Interpretability","level":2,"score":0.9398999810218811},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.79830002784729},{"id":"https://openalex.org/C52146309","wikidata":"https://www.wikidata.org/wiki/Q7431116","display_name":"Schema (genetic algorithms)","level":2,"score":0.5722000002861023},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4860999882221222},{"id":"https://openalex.org/C97931131","wikidata":"https://www.wikidata.org/wiki/Q5282087","display_name":"Discriminative model","level":2,"score":0.48260000348091125},{"id":"https://openalex.org/C2780513914","wikidata":"https://www.wikidata.org/wiki/Q18210350","display_name":"Bottleneck","level":2,"score":0.4787999987602234},{"id":"https://openalex.org/C66746571","wikidata":"https://www.wikidata.org/wiki/Q1134833","display_name":"ENCODE","level":3,"score":0.4740000069141388},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.4113999903202057},{"id":"https://openalex.org/C116409475","wikidata":"https://www.wikidata.org/wiki/Q1385056","display_name":"External Data Representation","level":2,"score":0.3384999930858612},{"id":"https://openalex.org/C21853045","wikidata":"https://www.wikidata.org/wiki/Q2634565","display_name":"Protocol data unit","level":3,"score":0.3325999975204468},{"id":"https://openalex.org/C118930307","wikidata":"https://www.wikidata.org/wiki/Q600590","display_name":"Tuple","level":2,"score":0.32829999923706055},{"id":"https://openalex.org/C2776359362","wikidata":"https://www.wikidata.org/wiki/Q2145286","display_name":"Representation (politics)","level":3,"score":0.32510000467300415},{"id":"https://openalex.org/C55439883","wikidata":"https://www.wikidata.org/wiki/Q360812","display_name":"Correctness","level":2,"score":0.3163999915122986},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.3140999972820282},{"id":"https://openalex.org/C2780385302","wikidata":"https://www.wikidata.org/wiki/Q367158","display_name":"Protocol (science)","level":3,"score":0.30889999866485596},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.30379998683929443},{"id":"https://openalex.org/C95623464","wikidata":"https://www.wikidata.org/wiki/Q1096149","display_name":"Classifier (UML)","level":2,"score":0.3028999865055084},{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.28949999809265137},{"id":"https://openalex.org/C205711294","wikidata":"https://www.wikidata.org/wiki/Q176953","display_name":"Rendering (computer graphics)","level":2,"score":0.2863999903202057},{"id":"https://openalex.org/C2776401178","wikidata":"https://www.wikidata.org/wiki/Q12050496","display_name":"Feature (linguistics)","level":2,"score":0.2827000021934509},{"id":"https://openalex.org/C162319229","wikidata":"https://www.wikidata.org/wiki/Q175263","display_name":"Data structure","level":2,"score":0.2786000072956085},{"id":"https://openalex.org/C2780719617","wikidata":"https://www.wikidata.org/wiki/Q1030752","display_name":"Salient","level":2,"score":0.2736999988555908},{"id":"https://openalex.org/C75608658","wikidata":"https://www.wikidata.org/wiki/Q44395","display_name":"Pascal (unit)","level":2,"score":0.2734000086784363},{"id":"https://openalex.org/C195324797","wikidata":"https://www.wikidata.org/wiki/Q33742","display_name":"Natural language","level":2,"score":0.25850000977516174},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.25119999051094055}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2607.00089","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2607.00089","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,"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.2607.00089","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2607.00089","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,"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":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Mechanistic":[0],"interpretability":[1,188],"has":[2],"produced":[3],"a":[4,69,80],"rich":[5],"inventory":[6],"of":[7],"component-level":[8],"analyses":[9,64],"that":[10,60,71],"characterise":[11],"what":[12],"neural-network":[13],"components":[14],"encode":[15],"and":[16,31,46,65,76,105,118,134,144,165],"how":[17],"they":[18],"interact.":[19],"Their":[20],"outputs,":[21],"however,":[22],"are":[23],"not":[24,41,47],"easily":[25],"reusable:":[26],"selectivity":[27],"tables,":[28],"circuit":[29,160],"diagrams,":[30],"feature":[32],"lists":[33],"remain":[34],"locked":[35],"in":[36,43],"per-study":[37],"notebooks":[38],"-":[39],"non-composable,":[40],"queryable":[42],"natural":[44],"language,":[45],"directly":[48],"actionable":[49],"for":[50,94,186],"downstream":[51,66],"audit":[52],"or":[53,177],"intervention.":[54],"We":[55,180],"study":[56],"the":[57,121,139],"representation":[58],"layer":[59],"sits":[61],"between":[62],"these":[63],"use":[67],"as":[68,183],"bottleneck":[70],"can":[72],"be":[73],"evaluated":[74],"independently,":[75],"introduce":[77],"Manifestation":[78],"Units,":[79],"typed":[81,126],"tuple":[82],"protocol":[83,122],"(E,":[84],"S,":[85],"R,":[86],"D,":[87],"G)":[88],"extended":[89],"with":[90,172],"attention-head":[91,153],"primitives":[92,154],"(T)":[93],"transformer":[95],"architectures,":[96],"organising":[97],"per-component":[98],"statistics":[99],"into":[100],"structured":[101],"fields":[102,174],"populated":[103],"automatically":[104],"queried":[106],"through":[107],"hybrid":[108],"retrieval.":[109],"Instantiated":[110],"across":[111],"generative":[112],"vision":[113,116],"(beta-VAE),":[114],"discriminative":[115],"(CNN),":[117],"language":[119],"(GPT-2),":[120],"supports":[123],"two":[124],"findings:":[125],"structure":[127],"substantially":[128],"outperforms":[129],"unstructured":[130],"baselines":[131],"on":[132],"retrieval,":[133],"CNN":[135],"filters":[136],"retrieved":[137],"by":[138],"schema":[140,151,184],"satisfy":[141],"causal":[142],"sufficiency":[143],"necessity":[145],"criteria":[146],"under":[147,162],"matched-budget":[148],"controls.":[149],"The":[150],"absorbs":[152],"without":[155],"modification,":[156],"set-recovers":[157],"known":[158],"IOI":[159],"members":[161],"retrieval-budget-matched":[163],"controls,":[164],"reveals":[166],"an":[167],"irreducible":[168],"two-field":[169],"core":[170],"(S+R)":[171],"remaining":[173],"either":[175],"redundant":[176],"actively":[178],"interfering.":[179],"present":[181],"this":[182],"infrastructure":[185],"mechanistic":[187],"rather":[189],"than":[190],"frontier-scale":[191],"validation.":[192]},"counts_by_year":[],"updated_date":"2026-07-28T07:46:37.118299","created_date":"2026-07-03T00:00:00"}