{"id":"https://openalex.org/W7163427062","doi":"https://doi.org/10.48550/arxiv.2606.03825","title":"Dynamic Short Convolutions Improve Transformers","display_name":"Dynamic Short Convolutions Improve Transformers","publication_year":2026,"publication_date":"2026-06-02","ids":{"openalex":"https://openalex.org/W7163427062","doi":"https://doi.org/10.48550/arxiv.2606.03825"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2606.03825","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.03825","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.2606.03825","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5093264931","display_name":"Oliver Sieberling","orcid":"https://orcid.org/0009-0008-4682-903X"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Sieberling, Oliver","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5000738825","display_name":"Bharat Runwal","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Runwal, Bharat","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5049734237","display_name":"Rameswar Panda","orcid":"https://orcid.org/0000-0003-4359-2475"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Panda, Rameswar","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5137741131","display_name":"Yoon Kim","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Kim, Yoon","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":0.5318999886512756,"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":0.5318999886512756,"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":0.09210000187158585,"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/T13702","display_name":"Machine Learning in Healthcare","score":0.04809999838471413,"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/transformer","display_name":"Transformer","score":0.597599983215332},{"id":"https://openalex.org/keywords/dynamic-range","display_name":"Dynamic range","score":0.5719000101089478},{"id":"https://openalex.org/keywords/scalability","display_name":"Scalability","score":0.5684999823570251},{"id":"https://openalex.org/keywords/high-dynamic-range","display_name":"High dynamic range","score":0.4706000089645386},{"id":"https://openalex.org/keywords/locality","display_name":"Locality","score":0.4586000144481659},{"id":"https://openalex.org/keywords/residual","display_name":"Residual","score":0.45179998874664307},{"id":"https://openalex.org/keywords/convolution","display_name":"Convolution (computer science)","score":0.39989998936653137}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7023000121116638},{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.597599983215332},{"id":"https://openalex.org/C87133666","wikidata":"https://www.wikidata.org/wiki/Q1161699","display_name":"Dynamic range","level":2,"score":0.5719000101089478},{"id":"https://openalex.org/C48044578","wikidata":"https://www.wikidata.org/wiki/Q727490","display_name":"Scalability","level":2,"score":0.5684999823570251},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.49300000071525574},{"id":"https://openalex.org/C2780056265","wikidata":"https://www.wikidata.org/wiki/Q106239881","display_name":"High dynamic range","level":3,"score":0.4706000089645386},{"id":"https://openalex.org/C2779808786","wikidata":"https://www.wikidata.org/wiki/Q6664603","display_name":"Locality","level":2,"score":0.4586000144481659},{"id":"https://openalex.org/C155512373","wikidata":"https://www.wikidata.org/wiki/Q287450","display_name":"Residual","level":2,"score":0.45179998874664307},{"id":"https://openalex.org/C45347329","wikidata":"https://www.wikidata.org/wiki/Q5166604","display_name":"Convolution (computer science)","level":3,"score":0.39989998936653137},{"id":"https://openalex.org/C118702147","wikidata":"https://www.wikidata.org/wiki/Q189396","display_name":"Dynamic random-access memory","level":3,"score":0.3057999908924103},{"id":"https://openalex.org/C99844830","wikidata":"https://www.wikidata.org/wiki/Q102441924","display_name":"Scaling","level":2,"score":0.3041999936103821},{"id":"https://openalex.org/C81363708","wikidata":"https://www.wikidata.org/wiki/Q17084460","display_name":"Convolutional neural network","level":2,"score":0.3019999861717224},{"id":"https://openalex.org/C24326235","wikidata":"https://www.wikidata.org/wiki/Q126095","display_name":"Electronic engineering","level":1,"score":0.2962999939918518},{"id":"https://openalex.org/C113775141","wikidata":"https://www.wikidata.org/wiki/Q428691","display_name":"Computer engineering","level":1,"score":0.2883000075817108},{"id":"https://openalex.org/C2780598303","wikidata":"https://www.wikidata.org/wiki/Q65921492","display_name":"Flexibility (engineering)","level":2,"score":0.28380000591278076},{"id":"https://openalex.org/C99580578","wikidata":"https://www.wikidata.org/wiki/Q5319002","display_name":"Dynamic problem","level":2,"score":0.27219998836517334},{"id":"https://openalex.org/C115051666","wikidata":"https://www.wikidata.org/wiki/Q6522493","display_name":"Ranging","level":2,"score":0.26409998536109924},{"id":"https://openalex.org/C159423971","wikidata":"https://www.wikidata.org/wiki/Q177251","display_name":"Associative property","level":2,"score":0.25769999623298645},{"id":"https://openalex.org/C94375191","wikidata":"https://www.wikidata.org/wiki/Q11205","display_name":"Arithmetic","level":1,"score":0.250900000333786}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2606.03825","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.03825","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.2606.03825","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.03825","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":[{"id":"https://metadata.un.org/sdg/4","score":0.41050785779953003,"display_name":"Quality Education"}],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Transformers":[0,99,101,117],"have":[1],"become":[2],"the":[3,13,51,124],"dominant":[4],"architecture":[5],"for":[6,37,187],"large":[7],"language":[8,190],"models,":[9],"largely":[10],"due":[11],"to":[12,67,91,123],"scalability":[14],"and":[15,23,70,100,127,130,152,184],"flexibility":[16],"of":[17,54],"attention,":[18],"feed-forward":[19],"layers,":[20],"residual":[21],"connections,":[22],"normalization.":[24],"This":[25],"paper":[26],"introduces":[27],"dynamic":[28,44,64,94,119,136,177],"short":[29,42,65,105,178],"convolutions":[30,45,66,95,120,137,143,179],"as":[31],"an":[32],"additional":[33],"neural":[34],"network":[35],"primitive":[36,186],"improving":[38],"Transformers.":[39],"Unlike":[40],"static":[41,82,104],"convolutions,":[43],"use":[46],"input-dependent":[47],"filters,":[48],"which":[49],"preserves":[50],"locality":[52],"bias":[53],"convolution":[55],"while":[56],"increasing":[57],"expressivity.":[58],"Motivating":[59],"experiments":[60,87],"show":[61],"that":[62,164,176],"applying":[63],"key,":[68,125],"query,":[69,126],"value":[71,128],"representations":[72],"improves":[73],"performance":[74],"on":[75,147],"challenging":[76],"associative":[77],"recall":[78],"tasks":[79],"compared":[80],"with":[81,103,160,168],"convolutional":[83],"variants.":[84],"Across":[85],"language-modeling":[86],"ranging":[88],"from":[89],"150M":[90],"2B":[92],"parameters,":[93],"consistently":[96],"outperform":[97],"standard":[98],"augmented":[102],"convolutions.":[106],"Fitting":[107],"scaling":[108],"laws":[109],"indicates":[110],"a":[111,131,169,181],"1.33$\\times$":[112],"compute":[113],"advantage":[114,133],"over":[115],"compute-matched":[116],"when":[118,134],"are":[121,180],"applied":[122],"vectors,":[129],"1.60$\\times$":[132],"adding":[135],"after":[138],"every":[139],"linear":[140,148],"layer.":[141],"Dynamic":[142],"also":[144],"offer":[145],"improvements":[146],"RNNs":[149],"(Mamba-2/Gated":[150],"DeltaNet)":[151],"mixture-of-experts":[153],"architectures.":[154],"We":[155],"make":[156],"these":[157],"gains":[158],"practical":[159],"custom":[161],"Triton":[162],"kernels":[163],"enable":[165],"efficient":[166],"training":[167],"manageable":[170],"end-to-end":[171],"slowdown.":[172],"These":[173],"results":[174],"suggest":[175],"scalable,":[182],"hardware-efficient,":[183],"expressive":[185],"advancing":[188],"Transformer-based":[189],"models.":[191]},"counts_by_year":[],"updated_date":"2026-07-28T07:46:37.118299","created_date":"2026-06-04T00:00:00"}