{"id":"https://openalex.org/W7163664995","doi":"https://doi.org/10.48550/arxiv.2606.06418","title":"Double Preconditioning (DoPr): Optimization for Test-Time Performance, not Validation Loss","display_name":"Double Preconditioning (DoPr): Optimization for Test-Time Performance, not Validation Loss","publication_year":2026,"publication_date":"2026-06-04","ids":{"openalex":"https://openalex.org/W7163664995","doi":"https://doi.org/10.48550/arxiv.2606.06418"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2606.06418","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.06418","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.2606.06418","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5019848297","display_name":"Thomas T. Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Thomas T.","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137969193","display_name":"Alok Shah","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Shah, Alok","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137991515","display_name":"Yifei Zhang (122595)","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Yifei","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137984347","display_name":"Vincent Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Vincent","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5052941508","display_name":"Nikolai Matni","orcid":"https://orcid.org/0000-0003-4936-3921"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Matni, Nikolai","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5037154191","display_name":"Max Simchowitz","orcid":"https://orcid.org/0000-0001-9900-1238"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Simchowitz, Max","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":0.2955999970436096,"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":0.2955999970436096,"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/T10036","display_name":"Advanced Neural Network Applications","score":0.13220000267028809,"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/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.1257999986410141,"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/task","display_name":"Task (project management)","score":0.6578999757766724},{"id":"https://openalex.org/keywords/key","display_name":"Key (lock)","score":0.6001999974250793},{"id":"https://openalex.org/keywords/downstream","display_name":"Downstream (manufacturing)","score":0.5663999915122986},{"id":"https://openalex.org/keywords/artificial-neural-network","display_name":"Artificial neural network","score":0.4943999946117401},{"id":"https://openalex.org/keywords/autoregressive-model","display_name":"Autoregressive model","score":0.486299991607666},{"id":"https://openalex.org/keywords/deep-learning","display_name":"Deep learning","score":0.44670000672340393},{"id":"https://openalex.org/keywords/range","display_name":"Range (aeronautics)","score":0.38089999556541443}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6934000253677368},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.6578999757766724},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.6001999974250793},{"id":"https://openalex.org/C2776207758","wikidata":"https://www.wikidata.org/wiki/Q5303302","display_name":"Downstream (manufacturing)","level":2,"score":0.5663999915122986},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5023999810218811},{"id":"https://openalex.org/C50644808","wikidata":"https://www.wikidata.org/wiki/Q192776","display_name":"Artificial neural network","level":2,"score":0.4943999946117401},{"id":"https://openalex.org/C159877910","wikidata":"https://www.wikidata.org/wiki/Q2202883","display_name":"Autoregressive model","level":2,"score":0.486299991607666},{"id":"https://openalex.org/C108583219","wikidata":"https://www.wikidata.org/wiki/Q197536","display_name":"Deep learning","level":2,"score":0.44670000672340393},{"id":"https://openalex.org/C204323151","wikidata":"https://www.wikidata.org/wiki/Q905424","display_name":"Range (aeronautics)","level":2,"score":0.38089999556541443},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.3781000077724457},{"id":"https://openalex.org/C175154964","wikidata":"https://www.wikidata.org/wiki/Q380077","display_name":"Task analysis","level":3,"score":0.32190001010894775},{"id":"https://openalex.org/C137836250","wikidata":"https://www.wikidata.org/wiki/Q984063","display_name":"Optimization problem","level":2,"score":0.3091999888420105},{"id":"https://openalex.org/C39890363","wikidata":"https://www.wikidata.org/wiki/Q36108","display_name":"Generative grammar","level":2,"score":0.29919999837875366},{"id":"https://openalex.org/C2778915421","wikidata":"https://www.wikidata.org/wiki/Q3643177","display_name":"Performance improvement","level":2,"score":0.28029999136924744},{"id":"https://openalex.org/C203274722","wikidata":"https://www.wikidata.org/wiki/Q7001161","display_name":"Network performance","level":2,"score":0.2703000009059906},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.267300009727478},{"id":"https://openalex.org/C79403827","wikidata":"https://www.wikidata.org/wiki/Q3988","display_name":"Real-time computing","level":1,"score":0.257999986410141},{"id":"https://openalex.org/C90509273","wikidata":"https://www.wikidata.org/wiki/Q11012","display_name":"Robot","level":2,"score":0.25589999556541443},{"id":"https://openalex.org/C2780440489","wikidata":"https://www.wikidata.org/wiki/Q5227278","display_name":"Data-driven","level":2,"score":0.251800000667572}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2606.06418","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.06418","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.2606.06418","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.06418","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":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Many":[0],"modern":[1],"applications":[2],"of":[3,67,127,151,165],"deep":[4],"learning":[5],"involve":[6],"training":[7],"a":[8,12,51,104,115,154,163],"neural":[9],"network":[10,23],"via":[11],"one-step":[13,193],"prediction":[14],"loss":[15,63],"(e.g.,":[16],"$L^2$":[17],"regression,":[18],"cross-entropy),":[19],"but":[20],"deploy":[21],"the":[22,58,61,125,149],"by":[24],"rolling":[25],"out":[26],"along":[27],"its":[28],"own":[29],"predictions.":[30],"Key":[31],"examples":[32],"include":[33],"autoregressive":[34],"language":[35],"modeling,":[36,39],"flow-based":[37],"generative":[38],"and":[40,64,74,86,136],"robot":[41],"policy":[42],"learning.":[43],"It":[44],"is":[45],"well-documented":[46],"that":[47,148],"these":[48,169],"settings":[49],"induce":[50],"phenomenon":[52],"we":[53,113],"call":[54],"test-time":[55,172],"feedback":[56],"(TTF):":[57],"mismatch":[59],"between":[60],"training/validation":[62],"downstream":[65,159],"metrics":[66],"interest,":[68],"such":[69,142],"as":[70,103,133,143],"task":[71,80],"success":[72],"rate":[73],"generation":[75],"quality,":[76],"which":[77],"grows":[78],"with":[79,138,192],"length.":[81],"While":[82],"data":[83],"curation,":[84],"architecture,":[85],"objective":[87],"design":[88,106],"have":[89],"been":[90],"proposed":[91],"to":[92,108,124,187],"combat":[93],"train-test":[94],"shift":[95],"in":[96,134,144,171,179],"TTF":[97,166],"settings,":[98],"this":[99],"paper":[100],"proposes":[101],"optimization":[102,117],"new":[105,116,183],"axis":[107],"mitigate":[109],"error":[110],"accumulation.":[111],"Specifically,":[112],"introduce":[114],"paradigm":[118],"called":[119],"double-preconditioning":[120],"(DoPr)":[121],"uniquely":[122],"tailored":[123],"challenges":[126],"TTF.":[128],"DoPr":[129],"combines":[130],"gradient-wise":[131],"preconditioning,":[132],"Adam":[135],"Muon,":[137],"activation-wise":[139],"preconditioning":[140],"(AP),":[141],"KFAC.":[145],"We":[146],"show":[147],"addition":[150],"AP":[152],"yields":[153],"drop-in":[155],"intervention":[156],"for":[157],"increasing":[158],"model":[160],"performance":[161,173],"across":[162],"range":[164],"settings.":[167],"Interestingly,":[168],"gains":[170],"do":[174],"not":[175],"consistently":[176],"accompany":[177],"improvements":[178],"validation":[180],"loss,":[181],"opening":[182],"questions":[184],"about":[185],"how":[186],"properly":[188],"evaluate":[189],"models":[190],"trained":[191],"supervised":[194],"objectives.":[195]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-06-06T00:00:00"}