{"id":"https://openalex.org/W7148963551","doi":"https://doi.org/10.48550/arxiv.2604.01411","title":"Test-Time Scaling Makes Overtraining Compute-Optimal","display_name":"Test-Time Scaling Makes Overtraining Compute-Optimal","publication_year":2026,"publication_date":"2026-04-01","ids":{"openalex":"https://openalex.org/W7148963551","doi":"https://doi.org/10.48550/arxiv.2604.01411"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2604.01411","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.01411","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":"article"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2604.01411","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5132868895","display_name":"Nicholas Roberts","orcid":null},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Roberts, Nicholas","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5102903741","display_name":"Sungjun Cho","orcid":"https://orcid.org/0000-0002-8609-6183"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Cho, Sungjun","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5132891134","display_name":"Zhiqi Gao","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Gao, Zhiqi","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5102608957","display_name":"Tzu-Heng Huang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Huang, Tzu-Heng","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5132835513","display_name":"Albert Wu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wu, Albert","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5001884614","display_name":"Gabriel Orlanski","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Orlanski, Gabriel","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5113392172","display_name":"Avi Trost","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Trost, Avi","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5006530776","display_name":"Kelly Buchanan","orcid":"https://orcid.org/0000-0002-7819-0583"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Buchanan, Kelly","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5050593520","display_name":"Aws Albarghouthi","orcid":"https://orcid.org/0000-0003-4577-175X"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Albarghouthi, Aws","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5103088414","display_name":"Fr\u00e9d\u00e9ric Sala","orcid":"https://orcid.org/0000-0003-0379-2827"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Sala, Frederic","raw_affiliation_strings":[],"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":10,"corresponding_author_ids":["https://openalex.org/A5132868895"],"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/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.17870000004768372,"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/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.17870000004768372,"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/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.11400000005960464,"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/T11612","display_name":"Stochastic Gradient Optimization Techniques","score":0.08410000056028366,"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/scaling","display_name":"Scaling","score":0.8058000206947327},{"id":"https://openalex.org/keywords/overtraining","display_name":"Overtraining","score":0.7996000051498413},{"id":"https://openalex.org/keywords/task","display_name":"Task (project management)","score":0.7329000234603882},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.6632999777793884},{"id":"https://openalex.org/keywords/scaling-law","display_name":"Scaling law","score":0.47510001063346863},{"id":"https://openalex.org/keywords/scale","display_name":"Scale (ratio)","score":0.4422000050544739}],"concepts":[{"id":"https://openalex.org/C99844830","wikidata":"https://www.wikidata.org/wiki/Q102441924","display_name":"Scaling","level":2,"score":0.8058000206947327},{"id":"https://openalex.org/C2778534509","wikidata":"https://www.wikidata.org/wiki/Q334310","display_name":"Overtraining","level":3,"score":0.7996000051498413},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.7329000234603882},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.6632999777793884},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.4844000041484833},{"id":"https://openalex.org/C2988430800","wikidata":"https://www.wikidata.org/wiki/Q428971","display_name":"Scaling law","level":3,"score":0.47510001063346863},{"id":"https://openalex.org/C2778755073","wikidata":"https://www.wikidata.org/wiki/Q10858537","display_name":"Scale (ratio)","level":2,"score":0.4422000050544739},{"id":"https://openalex.org/C185798385","wikidata":"https://www.wikidata.org/wiki/Q1161707","display_name":"Benchmark (surveying)","level":2,"score":0.37860000133514404},{"id":"https://openalex.org/C2778571376","wikidata":"https://www.wikidata.org/wiki/Q1355821","display_name":"Frontier","level":2,"score":0.35989999771118164},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.33809998631477356},{"id":"https://openalex.org/C204323151","wikidata":"https://www.wikidata.org/wiki/Q905424","display_name":"Range (aeronautics)","level":2,"score":0.3208000063896179},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.31520000100135803},{"id":"https://openalex.org/C149782125","wikidata":"https://www.wikidata.org/wiki/Q160039","display_name":"Econometrics","level":1,"score":0.3068000078201294},{"id":"https://openalex.org/C175154964","wikidata":"https://www.wikidata.org/wiki/Q380077","display_name":"Task analysis","level":3,"score":0.3041999936103821},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.30399999022483826},{"id":"https://openalex.org/C180747234","wikidata":"https://www.wikidata.org/wiki/Q23373","display_name":"Cognitive psychology","level":1,"score":0.29989999532699585}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2604.01411","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.01411","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":"article"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2604.01411","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.01411","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":"article"},"sustainable_development_goals":[{"id":"https://metadata.un.org/sdg/17","display_name":"Partnerships for the goals","score":0.4189230799674988}],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Modern":[0],"LLMs":[1,159],"scale":[2],"at":[3],"test-time,":[4],"e.g.":[5],"via":[6],"repeated":[7],"sampling,":[8],"where":[9],"inference":[10,51,109],"cost":[11],"grows":[12],"with":[13,62],"model":[14,44],"size":[15],"and":[16,48,73,93],"the":[17,90,117,122,139,168],"number":[18,49],"of":[19,50,121,124],"samples.":[20],"This":[21],"creates":[22],"a":[23],"trade-off":[24],"that":[25,41,105,142,164],"pretraining":[26,59,72,112,126,134,153],"scaling":[27,39,60,87,127,144,154,173],"laws,":[28],"such":[29],"as":[30,157],"Chinchilla,":[31],"do":[32],"not":[33],"address.":[34],"We":[35,129],"present":[36],"Train-to-Test":[37],"($T^2$)":[38],"laws":[40,61],"jointly":[42,70],"optimize":[43],"size,":[45],"training":[46],"tokens,":[47],"samples":[52],"under":[53],"fixed":[54],"end-to-end":[55],"budgets.":[56],"$T^2$":[57,78,143,172],"modernizes":[58],"pass@$k$":[63],"modeling":[64,83,94],"used":[65],"for":[66,108],"test-time":[67,74],"scaling,":[68],"then":[69],"optimizes":[71],"decisions.":[75],"Forecasts":[76],"from":[77],"are":[79,160],"robust":[80],"over":[81],"distinct":[82],"approaches:":[84],"measuring":[85],"joint":[86],"effect":[88],"on":[89,96],"task":[91,97],"loss":[92],"impact":[95],"accuracy.":[98],"Across":[99],"eight":[100],"downstream":[101],"tasks,":[102],"we":[103,162],"find":[104],"when":[106],"accounting":[107],"cost,":[110],"optimal":[111,140],"decisions":[113],"shift":[114],"radically":[115],"into":[116],"overtraining":[118],"regime,":[119],"well-outside":[120],"range":[123],"standard":[125],"suites.":[128],"validate":[130],"our":[131,165],"results":[132],"by":[133],"heavily":[135],"overtrained":[136],"models":[137],"in":[138,175],"region":[141],"forecasts,":[145],"confirming":[146],"their":[147],"substantially":[148],"stronger":[149],"performance":[150],"compared":[151],"to":[152],"alone.":[155],"Finally,":[156],"frontier":[158],"post-trained,":[161],"show":[163],"findings":[166],"survive":[167],"post-training":[169],"stage,":[170],"making":[171],"meaningful":[174],"modern":[176],"deployments.":[177]},"counts_by_year":[],"updated_date":"2026-04-04T06:15:33.020886","created_date":"2026-04-04T00:00:00"}