{"id":"https://openalex.org/W7148839664","doi":"https://doi.org/10.48550/arxiv.2604.01430","title":"Improving Latent Generalization Using Test-time Compute","display_name":"Improving Latent Generalization Using Test-time Compute","publication_year":2026,"publication_date":"2026-04-01","ids":{"openalex":"https://openalex.org/W7148839664","doi":"https://doi.org/10.48550/arxiv.2604.01430"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2604.01430","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.01430","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":"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.2604.01430","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5038288567","display_name":"Arslan Chaudhry","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chaudhry, Arslan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5104000581","display_name":"Sridhar Thiagarajan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Thiagarajan, Sridhar","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5132891149","display_name":"Andrew Lampinen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lampinen, Andrew","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/T10028","display_name":"Topic Modeling","score":0.5317000150680542,"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/T10028","display_name":"Topic Modeling","score":0.5317000150680542,"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/T11714","display_name":"Multimodal Machine Learning Applications","score":0.10419999808073044,"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":0.08100000023841858,"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/generalization","display_name":"Generalization","score":0.829800009727478},{"id":"https://openalex.org/keywords/correctness","display_name":"Correctness","score":0.5189999938011169},{"id":"https://openalex.org/keywords/latent-variable","display_name":"Latent variable","score":0.40299999713897705},{"id":"https://openalex.org/keywords/generalization-error","display_name":"Generalization error","score":0.3303999900817871},{"id":"https://openalex.org/keywords/reinforcement-learning","display_name":"Reinforcement learning","score":0.32679998874664307},{"id":"https://openalex.org/keywords/instance-based-learning","display_name":"Instance-based learning","score":0.27889999747276306}],"concepts":[{"id":"https://openalex.org/C177148314","wikidata":"https://www.wikidata.org/wiki/Q170084","display_name":"Generalization","level":2,"score":0.829800009727478},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6538000106811523},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.597599983215332},{"id":"https://openalex.org/C55439883","wikidata":"https://www.wikidata.org/wiki/Q360812","display_name":"Correctness","level":2,"score":0.5189999938011169},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.5062999725341797},{"id":"https://openalex.org/C51167844","wikidata":"https://www.wikidata.org/wiki/Q4422623","display_name":"Latent variable","level":2,"score":0.40299999713897705},{"id":"https://openalex.org/C117765406","wikidata":"https://www.wikidata.org/wiki/Q5362437","display_name":"Generalization error","level":3,"score":0.3303999900817871},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.32679998874664307},{"id":"https://openalex.org/C24138899","wikidata":"https://www.wikidata.org/wiki/Q17141258","display_name":"Instance-based learning","level":3,"score":0.27889999747276306},{"id":"https://openalex.org/C40506919","wikidata":"https://www.wikidata.org/wiki/Q7452469","display_name":"Sequence learning","level":2,"score":0.27410000562667847},{"id":"https://openalex.org/C67758918","wikidata":"https://www.wikidata.org/wiki/Q2744408","display_name":"Latent inhibition","level":4,"score":0.26589998602867126},{"id":"https://openalex.org/C65965080","wikidata":"https://www.wikidata.org/wiki/Q1806885","display_name":"Latent variable model","level":3,"score":0.26510000228881836},{"id":"https://openalex.org/C48164120","wikidata":"https://www.wikidata.org/wiki/Q4491893","display_name":"Concept learning","level":2,"score":0.2630000114440918},{"id":"https://openalex.org/C2778755073","wikidata":"https://www.wikidata.org/wiki/Q10858537","display_name":"Scale (ratio)","level":2,"score":0.2513999938964844}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2604.01430","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.01430","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":"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.2604.01430","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.01430","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":"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":[{"score":0.5773729681968689,"display_name":"Quality Education","id":"https://metadata.un.org/sdg/4"}],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Language":[0],"Models":[1],"(LMs)":[2],"exhibit":[3],"two":[4,25],"distinct":[5],"mechanisms":[6],"for":[7,168,221,236],"knowledge":[8,158,167,187],"acquisition:":[9],"in-weights":[10,30,74],"learning":[11,21,31,62,220],"(i.e.,":[12],"encoding":[13],"information":[14],"within":[15],"the":[16,39,54,190,216,238],"model":[17],"weights)":[18],"and":[19,90,233],"in-context":[20,61,219],"(ICL).":[22],"Although":[23],"these":[24,84,99],"modes":[26],"offer":[27],"complementary":[28],"strengths,":[29],"frequently":[32],"struggles":[33],"to":[34,92,94,109,116,128,131,136,165,198],"facilitate":[35],"deductive":[36],"reasoning":[37],"over":[38],"internalized":[40],"knowledge.":[41,96],"We":[42,120],"characterize":[43],"this":[44,101,144,222],"limitation":[45],"as":[46,230],"a":[47,231],"deficit":[48],"in":[49],"latent":[50,66,71,118,138,153,239],"generalization,":[51],"of":[52,152,193,206,218,241],"which":[53,169],"reversal":[55,177],"curse":[56],"is":[57],"one":[58],"example.":[59],"Conversely,":[60],"demonstrates":[63],"highly":[64],"robust":[65],"generalization":[67,72,154,240],"capabilities.":[68],"To":[69,97],"improve":[70,117,137],"from":[73,125],"knowledge,":[75],"prior":[76],"approaches":[77],"rely":[78],"on":[79,156,175],"train-time":[80],"data":[81],"augmentation,":[82],"yet":[83],"techniques":[85],"are":[86],"task-specific,":[87],"scale":[88],"poorly,":[89],"fail":[91],"generalize":[93],"out-of-distribution":[95],"overcome":[98],"shortcomings,":[100],"work":[102],"studies":[103],"how":[104],"models":[105,130,195,211],"can":[106],"be":[107],"taught":[108],"use":[110,121],"test-time":[111,228],"compute,":[112],"or":[113],"'thinking',":[114],"specifically":[115],"generalization.":[119,139],"Reinforcement":[122],"Learning":[123],"(RL)":[124],"correctness":[126],"feedback":[127],"train":[129],"produce":[132],"long":[133],"chains-of-thought":[134],"(CoTs)":[135],"Our":[140],"experiments":[141],"show":[142],"that":[143,181],"thinking":[145,182,194,210,229],"approach":[146],"not":[147,184],"only":[148],"resolves":[149],"many":[150],"instances":[151],"failures":[155],"in-distribution":[157],"but":[159,189],"also,":[160],"unlike":[161],"augmentation":[162],"baselines,":[163],"generalizes":[164],"new":[166],"no":[170],"RL":[171],"was":[172],"performed.":[173],"Nevertheless,":[174],"pure":[176],"tasks,":[178],"we":[179],"find":[180],"does":[183],"unlock":[185],"direct":[186],"inversion,":[188],"generate-and-verify":[191],"ability":[192],"enables":[196],"them":[197],"get":[199],"well":[200,214],"above":[201],"chance":[202],"performance.":[203],"The":[204],"brittleness":[205],"factual":[207],"self-verification":[208],"means":[209],"still":[212],"remain":[213],"below":[215],"performance":[217],"task.":[223],"Overall,":[224],"our":[225],"results":[226],"establish":[227],"flexible":[232],"promising":[234],"direction":[235],"improving":[237],"LMs.":[242]},"counts_by_year":[],"updated_date":"2026-07-28T07:46:37.118299","created_date":"2026-04-04T00:00:00"}