{"id":"https://openalex.org/W7164946343","doi":"https://doi.org/10.48550/arxiv.2606.15796","title":"DifFRACT: Diffusion Feature Reconstruction and Attribution for Circuit Tracing","display_name":"DifFRACT: Diffusion Feature Reconstruction and Attribution for Circuit Tracing","publication_year":2026,"publication_date":"2026-06-14","ids":{"openalex":"https://openalex.org/W7164946343","doi":"https://doi.org/10.48550/arxiv.2606.15796"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2606.15796","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.15796","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.2606.15796","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5138195058","display_name":"Artyom Mazur","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Mazur, Artyom","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5133477758","display_name":"Nina Konovalova","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Konovalova, Nina","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5138691410","display_name":"Aibek Alanov","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Alanov, Aibek","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/T10775","display_name":"Generative Adversarial Networks and Image Synthesis","score":0.2054000049829483,"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"}},"topics":[{"id":"https://openalex.org/T10775","display_name":"Generative Adversarial Networks and Image Synthesis","score":0.2054000049829483,"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/T12808","display_name":"Ferroelectric and Negative Capacitance Devices","score":0.07249999791383743,"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/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.058400001376867294,"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":0.8065999746322632},{"id":"https://openalex.org/keywords/tracing","display_name":"Tracing","score":0.6772000193595886},{"id":"https://openalex.org/keywords/granularity","display_name":"Granularity","score":0.45910000801086426},{"id":"https://openalex.org/keywords/transformer","display_name":"Transformer","score":0.42100000381469727},{"id":"https://openalex.org/keywords/feature","display_name":"Feature (linguistics)","score":0.41999998688697815},{"id":"https://openalex.org/keywords/generative-grammar","display_name":"Generative grammar","score":0.41040000319480896},{"id":"https://openalex.org/keywords/computation","display_name":"Computation","score":0.4075999855995178},{"id":"https://openalex.org/keywords/security-token","display_name":"Security token","score":0.398499995470047},{"id":"https://openalex.org/keywords/pattern-recognition","display_name":"Pattern recognition (psychology)","score":0.3853999972343445}],"concepts":[{"id":"https://openalex.org/C2781067378","wikidata":"https://www.wikidata.org/wiki/Q17027399","display_name":"Interpretability","level":2,"score":0.8065999746322632},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7014999985694885},{"id":"https://openalex.org/C138673069","wikidata":"https://www.wikidata.org/wiki/Q322229","display_name":"Tracing","level":2,"score":0.6772000193595886},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5748999714851379},{"id":"https://openalex.org/C177774035","wikidata":"https://www.wikidata.org/wiki/Q1246948","display_name":"Granularity","level":2,"score":0.45910000801086426},{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.42100000381469727},{"id":"https://openalex.org/C2776401178","wikidata":"https://www.wikidata.org/wiki/Q12050496","display_name":"Feature (linguistics)","level":2,"score":0.41999998688697815},{"id":"https://openalex.org/C39890363","wikidata":"https://www.wikidata.org/wiki/Q36108","display_name":"Generative grammar","level":2,"score":0.41040000319480896},{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.4075999855995178},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.398499995470047},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.3853999972343445},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.37049999833106995},{"id":"https://openalex.org/C167966045","wikidata":"https://www.wikidata.org/wiki/Q5532625","display_name":"Generative model","level":3,"score":0.35440000891685486},{"id":"https://openalex.org/C50644808","wikidata":"https://www.wikidata.org/wiki/Q192776","display_name":"Artificial neural network","level":2,"score":0.31619998812675476},{"id":"https://openalex.org/C205711294","wikidata":"https://www.wikidata.org/wiki/Q176953","display_name":"Rendering (computer graphics)","level":2,"score":0.31290000677108765},{"id":"https://openalex.org/C158622935","wikidata":"https://www.wikidata.org/wiki/Q660848","display_name":"Nonlinear system","level":2,"score":0.311599999666214},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.3077999949455261},{"id":"https://openalex.org/C165696696","wikidata":"https://www.wikidata.org/wiki/Q11287","display_name":"Exploit","level":2,"score":0.3052999973297119},{"id":"https://openalex.org/C52622490","wikidata":"https://www.wikidata.org/wiki/Q1026626","display_name":"Feature extraction","level":2,"score":0.3050999939441681},{"id":"https://openalex.org/C108583219","wikidata":"https://www.wikidata.org/wiki/Q197536","display_name":"Deep learning","level":2,"score":0.2904999852180481},{"id":"https://openalex.org/C99498987","wikidata":"https://www.wikidata.org/wiki/Q2210247","display_name":"Noise (video)","level":3,"score":0.28679999709129333},{"id":"https://openalex.org/C2781122975","wikidata":"https://www.wikidata.org/wiki/Q16928266","display_name":"Semantic feature","level":2,"score":0.26750001311302185},{"id":"https://openalex.org/C184337299","wikidata":"https://www.wikidata.org/wiki/Q1437428","display_name":"Semantics (computer science)","level":2,"score":0.2644999921321869},{"id":"https://openalex.org/C63479239","wikidata":"https://www.wikidata.org/wiki/Q7353546","display_name":"Robustness (evolution)","level":3,"score":0.2572999894618988}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2606.15796","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.15796","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.2606.15796","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.15796","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],"seeks":[2],"to":[3,110],"explain":[4],"neural":[5],"network":[6],"behavior":[7,123],"by":[8],"decomposing":[9],"model":[10],"computations":[11],"into":[12],"interpretable":[13,84,147],"features":[14,85,93],"and":[15,53,56,96,134,144,170,174,189,208,215],"circuits.":[16,148],"While":[17],"transcoder-based":[18,107,199],"circuit":[19,108,200],"tracing":[20,109],"has":[21],"recently":[22],"enabled":[23],"detailed":[24],"causal":[25,176],"analyses":[26],"of":[27,76,124],"large":[28],"language":[29],"models,":[30],"multimodal":[31,111,217],"diffusion":[32,112,206],"transformers":[33,207],"for":[34,44,178,204,213],"image":[35,57],"generation":[36,180],"remain":[37],"comparatively":[38],"opaque.":[39],"We":[40,114],"still":[41],"lack":[42],"tools":[43],"understanding":[45,214],"how":[46,54,91],"semantic":[47,172],"information":[48],"propagates":[49],"across":[50],"denoising":[51],"steps":[52],"text":[55],"representations":[58],"interact":[59],"within":[60],"double-stream":[61],"MM-DiT":[62],"architectures.":[63],"Existing":[64],"methods":[65],"provide":[66,175],"only":[67],"partial":[68],"insight:":[69],"attention":[70],"maps":[71],"expose":[72],"a":[73,210],"limited":[74],"view":[75],"token":[77],"interactions,":[78],"while":[79],"sparse":[80,156],"autoencoders":[81,157],"can":[82],"discover":[83],"but":[86],"do":[87],"not":[88],"directly":[89],"reveal":[90,165],"these":[92],"are":[94,185],"transformed":[95],"composed":[97],"through":[98],"nonlinear":[99],"MLP":[100,125],"layers.":[101],"In":[102],"this":[103],"work,":[104],"we":[105,139],"extend":[106],"transformers.":[113],"train":[115],"timestep-conditioned":[116],"transcoders":[117,133,151],"that":[118,198],"faithfully":[119],"approximate":[120],"the":[121,136,159],"input-output":[122],"sublayers":[126],"in":[127],"FLUX.1[schnell].":[128],"By":[129],"replacing":[130],"MLPs":[131],"with":[132],"linearizing":[135],"remaining":[137],"computation,":[138],"obtain":[140],"exact":[141],"feature-to-feature":[142],"attribution":[143],"recover":[145],"compact,":[146],"Empirically,":[149],"our":[150],"match":[152],"or":[153],"slightly":[154],"outperform":[155],"on":[158],"sparsity-faithfulness":[160],"tradeoff.":[161],"The":[162,220],"resulting":[163],"circuits":[164],"mechanisms":[166],"underlying":[167],"attribute":[168],"binding":[169],"cross-stream":[171],"propagation,":[173],"explanations":[177],"systematic":[179],"errors.":[181],"Moreover,":[182],"circuit-guided":[183],"interventions":[184],"substantially":[186],"more":[187],"precise":[188],"effective":[190],"than":[191],"standard":[192],"SAE-based":[193],"steering.":[194],"Our":[195],"results":[196],"demonstrate":[197],"analysis":[201],"is":[202,222],"feasible":[203],"state-of-the-art":[205],"provides":[209],"powerful":[211],"framework":[212],"controlling":[216],"generative":[218],"models.":[219],"code":[221],"available":[223],"at":[224],"https://github.com/Artalmaz31/DifFRACT":[225]},"counts_by_year":[],"updated_date":"2026-07-28T07:46:37.118299","created_date":"2026-06-17T00:00:00"}