{"id":"https://openalex.org/W7166885035","doi":"https://doi.org/10.48550/arxiv.2606.31315","title":"BlockPilot: Instance-Adaptive Policy Learning for Diffusion-based Speculative Decoding","display_name":"BlockPilot: Instance-Adaptive Policy Learning for Diffusion-based Speculative Decoding","publication_year":2026,"publication_date":"2026-06-30","ids":{"openalex":"https://openalex.org/W7166885035","doi":"https://doi.org/10.48550/arxiv.2606.31315"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2606.31315","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.31315","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":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.2606.31315","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5139806161","display_name":"Hao Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Hao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5102712817","display_name":"Yiming Hu","orcid":"https://orcid.org/0009-0006-2418-976X"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Hu, Yiming","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139793598","display_name":"Yong Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Yong","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5111267280","display_name":"Mingqiao Mo","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Mo, Mingqiao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100559024","display_name":"Xin Xiao","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xiao, Xin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5056733554","display_name":"Xi Chu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chu, Xiangxiang","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":0.7748000025749207,"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":0.7748000025749207,"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/T10201","display_name":"Speech Recognition and Synthesis","score":0.030899999663233757,"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":0.013199999928474426,"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":0.8120999932289124},{"id":"https://openalex.org/keywords/block","display_name":"Block (permutation group theory)","score":0.6639999747276306},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.5512999892234802},{"id":"https://openalex.org/keywords/block-size","display_name":"Block size","score":0.5511999726295471},{"id":"https://openalex.org/keywords/speedup","display_name":"Speedup","score":0.5364999771118164},{"id":"https://openalex.org/keywords/selection","display_name":"Selection (genetic algorithm)","score":0.4966000020503998},{"id":"https://openalex.org/keywords/lossless-compression","display_name":"Lossless compression","score":0.4490000009536743}],"concepts":[{"id":"https://openalex.org/C57273362","wikidata":"https://www.wikidata.org/wiki/Q576722","display_name":"Decoding methods","level":2,"score":0.8120999932289124},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7538999915122986},{"id":"https://openalex.org/C2777210771","wikidata":"https://www.wikidata.org/wiki/Q4927124","display_name":"Block (permutation group theory)","level":2,"score":0.6639999747276306},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.5512999892234802},{"id":"https://openalex.org/C41431624","wikidata":"https://www.wikidata.org/wiki/Q1053357","display_name":"Block size","level":3,"score":0.5511999726295471},{"id":"https://openalex.org/C68339613","wikidata":"https://www.wikidata.org/wiki/Q1549489","display_name":"Speedup","level":2,"score":0.5364999771118164},{"id":"https://openalex.org/C81917197","wikidata":"https://www.wikidata.org/wiki/Q628760","display_name":"Selection (genetic algorithm)","level":2,"score":0.4966000020503998},{"id":"https://openalex.org/C81081738","wikidata":"https://www.wikidata.org/wiki/Q55542","display_name":"Lossless compression","level":3,"score":0.4490000009536743},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.42329999804496765},{"id":"https://openalex.org/C2776359362","wikidata":"https://www.wikidata.org/wiki/Q2145286","display_name":"Representation (politics)","level":3,"score":0.4205999970436096},{"id":"https://openalex.org/C157125643","wikidata":"https://www.wikidata.org/wiki/Q884707","display_name":"Block code","level":3,"score":0.3361999988555908},{"id":"https://openalex.org/C125411270","wikidata":"https://www.wikidata.org/wiki/Q18653","display_name":"Encoding (memory)","level":2,"score":0.33570000529289246},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.33399999141693115},{"id":"https://openalex.org/C22367795","wikidata":"https://www.wikidata.org/wiki/Q7625208","display_name":"Structured prediction","level":2,"score":0.33149999380111694},{"id":"https://openalex.org/C179799912","wikidata":"https://www.wikidata.org/wiki/Q205084","display_name":"Computational complexity theory","level":2,"score":0.3077000081539154},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3012999892234802},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.2628999948501587}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2606.31315","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.31315","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":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.2606.31315","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.31315","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"sustainable_development_goals":[{"id":"https://metadata.un.org/sdg/16","score":0.49318069219589233,"display_name":"Peace, Justice and strong institutions"}],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Speculative":[0],"decoding":[1,30,62,92],"accelerates":[2],"inference":[3,54],"by":[4,20,34],"using":[5],"a":[6,52,59,87,98,113,126,146,205],"lightweight":[7,147],"draft":[8],"model":[9],"to":[10,112],"generate":[11],"candidate":[12],"tokens":[13,37],"in":[14,90],"parallel,":[15],"and":[16,57,85,115,151,194,204],"are":[17],"then":[18],"verified":[19],"the":[21,78,104,110,131,136,159,165,168],"target":[22],"model,":[23],"enabling":[24],"lossless":[25],"acceleration.":[26],"Recently,":[27],"diffusion-based":[28],"speculative":[29,91],"further":[31],"improves":[32,196],"parallelism":[33],"generating":[35],"multiple":[36],"per":[38],"forward":[39],"pass":[40],"via":[41],"block-level":[42],"diffusion,":[43],"achieving":[44,198],"state-of-the-art":[45],"(SOTA)":[46],"performance.":[47,93],"However,":[48],"existing":[49],"methods":[50],"adopt":[51],"fixed":[53],"block":[55,80,106,133,142,161],"size":[56,81,134,143,162],"assume":[58],"uniform":[60],"optimal":[61,79,132,160],"strategy":[63],"across":[64,83],"all":[65],"inputs.":[66],"In":[67],"this":[68,73],"paper,":[69],"we":[70,123,140],"show":[71],"that":[72,129,157,186],"assumption":[74],"is":[75,173,189],"suboptimal,":[76],"as":[77,145],"varies":[82],"samples":[84],"plays":[86],"critical":[88],"role":[89],"Moreover,":[94],"these":[95,121],"values":[96],"exhibit":[97],"clear":[99],"local":[100],"structure,":[101],"concentrating":[102],"around":[103],"training":[105],"size,":[107],"which":[108],"reduces":[109],"problem":[111,150],"low-dimensional":[114],"structured":[116],"decision":[117,155],"space.":[118],"Based":[119],"on":[120,164,208],"insights,":[122],"propose":[124,152],"BlockPilot,":[125],"sample-adaptive":[127],"policy":[128,148],"predicts":[130,158],"from":[135],"prefilling":[137,169],"representation.":[138],"Specifically,":[139],"formulate":[141],"selection":[144],"learning":[149],"an":[153,199],"instance-adaptive":[154],"mechanism":[156],"based":[163],"representation":[166],"of":[167,202],"stage.":[170],"The":[171],"prediction":[172],"performed":[174],"only":[175],"once":[176],"after":[177],"prefilling,":[178],"allowing":[179],"for":[180],"seamless":[181],"integration.":[182],"Extensive":[183],"experiments":[184],"demonstrate":[185],"our":[187],"method":[188],"plug-and-play,":[190],"introduces":[191],"minimal":[192],"overhead,":[193],"consistently":[195],"efficiency,":[197],"acceptance":[200],"length":[201],"5.92":[203],"4.20$\\times$":[206],"speedup":[207],"Qwen3-4B":[209],"under":[210],"temperature":[211],"$T=1$.":[212]},"counts_by_year":[],"updated_date":"2026-07-02T06:18:51.028212","created_date":"2026-07-02T00:00:00"}