{"id":"https://openalex.org/W7151316791","doi":"https://doi.org/10.48550/arxiv.2604.03674","title":"DiffSparse: Accelerating Diffusion Transformers with Learned Token Sparsity","display_name":"DiffSparse: Accelerating Diffusion Transformers with Learned Token Sparsity","publication_year":2026,"publication_date":"2026-04-04","ids":{"openalex":"https://openalex.org/W7151316791","doi":"https://doi.org/10.48550/arxiv.2604.03674"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2604.03674","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.03674","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.2604.03674","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5133074180","display_name":"Haowei Zhu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhu, Haowei","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5133090176","display_name":"Ji Liu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liu, Ji","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5079691252","display_name":"Ziqiong Liu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liu, Ziqiong","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5133134357","display_name":"Dong Li","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Dong","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5133074641","display_name":"Junhai Yong","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yong, Junhai","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5133110275","display_name":"Bin Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Bin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5133104765","display_name":"Emad Barsoum","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Barsoum, Emad","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":0.6621000170707703,"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":0.6621000170707703,"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/T11304","display_name":"Advanced Neuroimaging Techniques and Applications","score":0.0892999991774559,"subfield":{"id":"https://openalex.org/subfields/2741","display_name":"Radiology, Nuclear Medicine and Imaging"},"field":{"id":"https://openalex.org/fields/27","display_name":"Medicine"},"domain":{"id":"https://openalex.org/domains/4","display_name":"Health Sciences"}},{"id":"https://openalex.org/T11019","display_name":"Image Enhancement Techniques","score":0.05849999934434891,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/security-token","display_name":"Security token","score":0.7343000173568726},{"id":"https://openalex.org/keywords/cache","display_name":"Cache","score":0.6517999768257141},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.59579998254776},{"id":"https://openalex.org/keywords/granularity","display_name":"Granularity","score":0.5758000016212463},{"id":"https://openalex.org/keywords/computation","display_name":"Computation","score":0.5685999989509583},{"id":"https://openalex.org/keywords/transformer","display_name":"Transformer","score":0.40529999136924744}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8445000052452087},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.7343000173568726},{"id":"https://openalex.org/C115537543","wikidata":"https://www.wikidata.org/wiki/Q165596","display_name":"Cache","level":2,"score":0.6517999768257141},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.59579998254776},{"id":"https://openalex.org/C177774035","wikidata":"https://www.wikidata.org/wiki/Q1246948","display_name":"Granularity","level":2,"score":0.5758000016212463},{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.5685999989509583},{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.40529999136924744},{"id":"https://openalex.org/C120314980","wikidata":"https://www.wikidata.org/wiki/Q180634","display_name":"Distributed computing","level":1,"score":0.38019999861717224},{"id":"https://openalex.org/C113775141","wikidata":"https://www.wikidata.org/wiki/Q428691","display_name":"Computer engineering","level":1,"score":0.32519999146461487},{"id":"https://openalex.org/C179799912","wikidata":"https://www.wikidata.org/wiki/Q205084","display_name":"Computational complexity theory","level":2,"score":0.31690001487731934},{"id":"https://openalex.org/C37404715","wikidata":"https://www.wikidata.org/wiki/Q380679","display_name":"Dynamic programming","level":2,"score":0.31130000948905945},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.2962999939918518},{"id":"https://openalex.org/C138236772","wikidata":"https://www.wikidata.org/wiki/Q25098575","display_name":"Edge device","level":3,"score":0.2953999936580658},{"id":"https://openalex.org/C79403827","wikidata":"https://www.wikidata.org/wiki/Q3988","display_name":"Real-time computing","level":1,"score":0.2851000130176544},{"id":"https://openalex.org/C204323151","wikidata":"https://www.wikidata.org/wiki/Q905424","display_name":"Range (aeronautics)","level":2,"score":0.2797999978065491},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.27160000801086426},{"id":"https://openalex.org/C2776401178","wikidata":"https://www.wikidata.org/wiki/Q12050496","display_name":"Feature (linguistics)","level":2,"score":0.26350000500679016},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.25929999351501465}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2604.03674","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.03674","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.2604.03674","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.03674","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":[],"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],"models":[1,45],"demonstrate":[2,196],"outstanding":[3],"performance":[4,40],"in":[5,41,64,67,105,131],"image":[6],"generation,":[7],"but":[8],"their":[9],"multi-step":[10],"inference":[11,20],"mechanism":[12],"requires":[13],"immense":[14],"computational":[15,30,175],"cost.":[16,31],"Previous":[17],"works":[18],"accelerate":[19],"by":[21,177],"leveraging":[22,88],"layer":[23],"or":[24],"token":[25,69,89,93],"cache":[26,70],"techniques":[27],"to":[28,36,47,91],"reduce":[29,92,174],"However,":[32],"these":[33,68,74,154],"methods":[34],"fail":[35],"achieve":[37],"superior":[38],"acceleration":[39],"few-step":[42],"diffusion":[43,85],"transformer":[44,86],"due":[46],"inefficient":[48],"feature":[49],"caching":[50,90],"strategies,":[51],"manually":[52],"designed":[53],"sparsity":[54,81,103],"allocation,":[55],"and":[56,96,151],"the":[57,126,187],"practice":[58],"of":[59,144,186],"retaining":[60],"complete":[61],"forward":[62],"computations":[63],"several":[65],"steps":[66],"methods.":[71],"To":[72],"tackle":[73],"challenges,":[75],"we":[76,173],"propose":[77],"a":[78,110,115,142],"differentiable":[79],"layer-wise":[80,102],"optimization":[82],"framework":[83],"for":[84,128],"models,":[87,146],"computation":[94],"costs":[95],"enhance":[97],"acceleration.":[98],"Our":[99],"method":[100,157,199],"optimizes":[101],"allocation":[104],"an":[106],"end-to-end":[107],"manner":[108],"through":[109],"learnable":[111],"network":[112],"combined":[113],"with":[114,169],"dynamic":[116],"programming":[117],"solver.":[118],"Additionally,":[119],"our":[120,156,198],"proposed":[121],"two-stage":[122],"training":[123],"strategy":[124],"eliminates":[125],"need":[127],"full-step":[129],"processing":[130],"existing":[132],"methods,":[133],"further":[134],"improving":[135,206],"efficiency.":[136],"We":[137],"conducted":[138],"extensive":[139],"experiments":[140],"on":[141,167],"range":[143],"diffusion-transformer":[145],"including":[147],"DiT-XL/2,":[148],"PixArt-$\u03b1$,":[149],"FLUX,":[150],"Wan2.1.":[152],"Across":[153],"architectures,":[155],"consistently":[158],"improves":[159],"efficiency":[160,202],"without":[161],"degrading":[162],"sample":[163],"quality.":[164,208],"For":[165],"example,":[166],"PixArt-$\u03b1$":[168],"20":[170],"sampling":[171],"steps,":[172],"cost":[176],"$54\\%$":[178],"while":[179,204],"achieving":[180],"generation":[181,207],"metrics":[182],"that":[183,197],"surpass":[184],"those":[185],"original":[188],"model,":[189],"substantially":[190],"outperforming":[191],"prior":[192],"approaches.":[193],"These":[194],"results":[195],"delivers":[200],"large":[201],"gains":[203],"often":[205]},"counts_by_year":[],"updated_date":"2026-07-28T07:46:37.118299","created_date":"2026-04-08T00:00:00"}