{"id":"https://openalex.org/W7140197266","doi":"https://doi.org/10.48550/arxiv.2603.22248","title":"Confidence-Based Decoding is Provably Efficient for Diffusion Language Models","display_name":"Confidence-Based Decoding is Provably Efficient for Diffusion Language Models","publication_year":2026,"publication_date":"2026-03-23","ids":{"openalex":"https://openalex.org/W7140197266","doi":"https://doi.org/10.48550/arxiv.2603.22248"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2603.22248","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.22248","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.2603.22248","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":null,"display_name":"Cai, Changxiao","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Cai, Changxiao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":null,"display_name":"Li, Gen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Gen","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":0.29649999737739563,"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":0.29649999737739563,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.21209999918937683,"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/T10201","display_name":"Speech Recognition and Synthesis","score":0.08190000057220459,"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/decoding-methods","display_name":"Decoding methods","score":0.7702000141143799},{"id":"https://openalex.org/keywords/entropy","display_name":"Entropy (arrow of time)","score":0.499099999666214},{"id":"https://openalex.org/keywords/importance-sampling","display_name":"Importance sampling","score":0.46810001134872437},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.454800009727478},{"id":"https://openalex.org/keywords/autoregressive-model","display_name":"Autoregressive model","score":0.43140000104904175},{"id":"https://openalex.org/keywords/principle-of-maximum-entropy","display_name":"Principle of maximum entropy","score":0.3912999927997589},{"id":"https://openalex.org/keywords/sampling","display_name":"Sampling (signal processing)","score":0.38119998574256897},{"id":"https://openalex.org/keywords/probability-distribution","display_name":"Probability distribution","score":0.37130001187324524},{"id":"https://openalex.org/keywords/kullback\u2013leibler-divergence","display_name":"Kullback\u2013Leibler divergence","score":0.35580000281333923}],"concepts":[{"id":"https://openalex.org/C57273362","wikidata":"https://www.wikidata.org/wiki/Q576722","display_name":"Decoding methods","level":2,"score":0.7702000141143799},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7164000272750854},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.499099999666214},{"id":"https://openalex.org/C106301342","wikidata":"https://www.wikidata.org/wiki/Q4117933","display_name":"Entropy (arrow of time)","level":2,"score":0.499099999666214},{"id":"https://openalex.org/C52740198","wikidata":"https://www.wikidata.org/wiki/Q1539564","display_name":"Importance sampling","level":3,"score":0.46810001134872437},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.454800009727478},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.4399999976158142},{"id":"https://openalex.org/C159877910","wikidata":"https://www.wikidata.org/wiki/Q2202883","display_name":"Autoregressive model","level":2,"score":0.43140000104904175},{"id":"https://openalex.org/C9679016","wikidata":"https://www.wikidata.org/wiki/Q1417473","display_name":"Principle of maximum entropy","level":2,"score":0.3912999927997589},{"id":"https://openalex.org/C140779682","wikidata":"https://www.wikidata.org/wiki/Q210868","display_name":"Sampling (signal processing)","level":3,"score":0.38119998574256897},{"id":"https://openalex.org/C149441793","wikidata":"https://www.wikidata.org/wiki/Q200726","display_name":"Probability distribution","level":2,"score":0.37130001187324524},{"id":"https://openalex.org/C171752962","wikidata":"https://www.wikidata.org/wiki/Q255166","display_name":"Kullback\u2013Leibler divergence","level":2,"score":0.35580000281333923},{"id":"https://openalex.org/C167981619","wikidata":"https://www.wikidata.org/wiki/Q1685498","display_name":"Cross entropy","level":3,"score":0.3391999900341034},{"id":"https://openalex.org/C52622258","wikidata":"https://www.wikidata.org/wiki/Q131222","display_name":"Information theory","level":2,"score":0.3248000144958496},{"id":"https://openalex.org/C192209626","wikidata":"https://www.wikidata.org/wiki/Q190909","display_name":"Focus (optics)","level":2,"score":0.3005000054836273},{"id":"https://openalex.org/C204397858","wikidata":"https://www.wikidata.org/wiki/Q4437907","display_name":"List decoding","level":5,"score":0.2955000102519989},{"id":"https://openalex.org/C75291252","wikidata":"https://www.wikidata.org/wiki/Q1315756","display_name":"TRACE (psycholinguistics)","level":2,"score":0.2948000133037567},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.29350000619888306},{"id":"https://openalex.org/C98385598","wikidata":"https://www.wikidata.org/wiki/Q1339385","display_name":"Empirical distribution function","level":2,"score":0.2912999987602234},{"id":"https://openalex.org/C179518139","wikidata":"https://www.wikidata.org/wiki/Q5140297","display_name":"Coding (social sciences)","level":2,"score":0.29089999198913574},{"id":"https://openalex.org/C124551494","wikidata":"https://www.wikidata.org/wiki/Q3055345","display_name":"Differential entropy","level":4,"score":0.2858999967575073},{"id":"https://openalex.org/C14828074","wikidata":"https://www.wikidata.org/wiki/Q186559","display_name":"Arrow of time","level":3,"score":0.26269999146461487},{"id":"https://openalex.org/C29406490","wikidata":"https://www.wikidata.org/wiki/Q1420659","display_name":"Fisher information","level":2,"score":0.259799987077713},{"id":"https://openalex.org/C207390915","wikidata":"https://www.wikidata.org/wiki/Q1230525","display_name":"Divergence (linguistics)","level":2,"score":0.2572999894618988},{"id":"https://openalex.org/C73602740","wikidata":"https://www.wikidata.org/wiki/Q7795822","display_name":"Thompson sampling","level":3,"score":0.2515999972820282}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2603.22248","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.22248","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.2603.22248","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.22248","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":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Diffusion":[0],"language":[1,15],"models":[2,13],"(DLMs)":[3],"have":[4,80],"emerged":[5],"as":[6],"a":[7,31,130,199],"promising":[8],"alternative":[9],"to":[10,74,175,182],"autoregressive":[11],"(AR)":[12],"for":[14,106,202,215],"modeling,":[16],"allowing":[17],"flexible":[18],"generation":[19,23],"order":[20,44],"and":[21,45,70,132,205],"parallel":[22],"of":[24,47,91,146,155,186,210],"multiple":[25],"tokens.":[26],"However,":[27],"this":[28,86,97,161],"flexibility":[29],"introduces":[30],"challenge":[32],"absent":[33],"in":[34,62,109,139],"AR":[35],"models:":[36],"the":[37,43,101,126,153,156,168,176,183,208],"\\emph{decoding":[38],"strategy}":[39],"--":[40,53],"which":[41,66,69],"determines":[42],"number":[46,145],"tokens":[48,73,121],"generated":[49],"at":[50],"each":[51,123],"iteration":[52,124],"critically":[54],"affects":[55],"sampling":[56,138,165],"efficiency.":[57],"Among":[58],"decoding":[59,93,108,204,213],"strategies":[60,214],"explored":[61],"practice,":[63],"confidence-based":[64,92,107,203],"methods,":[65],"adaptively":[67],"select":[68],"how":[71],"many":[72],"unmask":[75],"based":[76],"on":[77,113],"prediction":[78],"confidence,":[79],"shown":[81],"strong":[82],"empirical":[83],"performance.":[84],"Despite":[85],"success,":[87],"our":[88,196],"theoretical":[89,103,200],"understanding":[90],"remains":[94],"limited.":[95],"In":[96],"work,":[98],"we":[99],"develop":[100],"first":[102],"analysis":[104],"framework":[105],"DLMs.":[110,216],"We":[111],"focus":[112],"an":[114,143],"entropy":[115,128,154,173],"sum-based":[116],"strategy":[117,162],"that":[118,134],"continues":[119],"unmasking":[120],"within":[122],"until":[125],"cumulative":[127],"exceeds":[129],"threshold,":[131],"show":[133],"it":[135],"achieves":[136],"$\\varepsilon$-accurate":[137],"KL":[140],"divergence":[141],"with":[142],"expected":[144],"iterations":[147],"$\\widetilde":[148],"O(H(X_0)/\\varepsilon)$,":[149],"where":[150],"$H(X_0)$":[151],"denotes":[152],"target":[157],"data":[158,169,187],"distribution.":[159],"Notably,":[160],"yields":[163],"substantial":[164],"acceleration":[166],"when":[167],"distribution":[170],"has":[171],"low":[172],"relative":[174],"sequence":[177],"length,":[178],"while":[179],"automatically":[180],"adapting":[181],"intrinsic":[184],"complexity":[185],"without":[188],"requiring":[189],"prior":[190],"knowledge":[191],"or":[192],"hyperparameter":[193],"tuning.":[194],"Overall,":[195],"results":[197],"provide":[198],"foundation":[201],"may":[206],"inform":[207],"design":[209],"more":[211],"efficient":[212]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-03-25T00:00:00"}