{"id":"https://openalex.org/W7163059097","doi":"https://doi.org/10.48550/arxiv.2605.30523","title":"Revisiting Padded Transformer Expressivity: Which Architectural Choices Matter and Which Don't","display_name":"Revisiting Padded Transformer Expressivity: Which Architectural Choices Matter and Which Don't","publication_year":2026,"publication_date":"2026-05-28","ids":{"openalex":"https://openalex.org/W7163059097","doi":"https://doi.org/10.48550/arxiv.2605.30523"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.30523","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.30523","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":"article"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2605.30523","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5071101707","display_name":"Anej Svete","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Svete, Anej","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5125553375","display_name":"William Merrill","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Merrill, William","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5061951606","display_name":"Ryan Cotterell","orcid":"https://orcid.org/0000-0003-4080-1833"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Cotterell, Ryan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5077726785","display_name":"Ashish Sabharwal","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Sabharwal, Ashish","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":4,"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/T12808","display_name":"Ferroelectric and Negative Capacitance Devices","score":0.19939999282360077,"subfield":{"id":"https://openalex.org/subfields/2208","display_name":"Electrical and Electronic Engineering"},"field":{"id":"https://openalex.org/fields/22","display_name":"Engineering"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T12808","display_name":"Ferroelectric and Negative Capacitance Devices","score":0.19939999282360077,"subfield":{"id":"https://openalex.org/subfields/2208","display_name":"Electrical and Electronic Engineering"},"field":{"id":"https://openalex.org/fields/22","display_name":"Engineering"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.18930000066757202,"subfield":{"id":"https://openalex.org/subfields/1708","display_name":"Hardware and Architecture"},"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/T10036","display_name":"Advanced Neural Network Applications","score":0.12460000067949295,"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/transformer","display_name":"Transformer","score":0.7333999872207642},{"id":"https://openalex.org/keywords/softmax-function","display_name":"Softmax function","score":0.4456000030040741},{"id":"https://openalex.org/keywords/logarithm","display_name":"Logarithm","score":0.382099986076355},{"id":"https://openalex.org/keywords/gadget","display_name":"Gadget","score":0.23800000548362732}],"concepts":[{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.7333999872207642},{"id":"https://openalex.org/C188441871","wikidata":"https://www.wikidata.org/wiki/Q7554146","display_name":"Softmax function","level":3,"score":0.4456000030040741},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.43709999322891235},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.41990000009536743},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.3921000063419342},{"id":"https://openalex.org/C39927690","wikidata":"https://www.wikidata.org/wiki/Q11197","display_name":"Logarithm","level":2,"score":0.382099986076355},{"id":"https://openalex.org/C119770614","wikidata":"https://www.wikidata.org/wiki/Q5516347","display_name":"Gadget","level":2,"score":0.23800000548362732},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.2362000048160553},{"id":"https://openalex.org/C94375191","wikidata":"https://www.wikidata.org/wiki/Q11205","display_name":"Arithmetic","level":1,"score":0.23589999973773956},{"id":"https://openalex.org/C199639397","wikidata":"https://www.wikidata.org/wiki/Q1788588","display_name":"Engineering drawing","level":1,"score":0.23260000348091125}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.30523","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.30523","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":"article"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2605.30523","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.30523","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":"article"},"sustainable_development_goals":[{"display_name":"Sustainable cities and communities","score":0.47105470299720764,"id":"https://metadata.un.org/sdg/11"}],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Recent":[0],"work":[1],"describes":[2],"what":[3],"transformers":[4,27,94,123,149],"can":[5],"and":[6,20,85,102,106,153,170,178],"cannot":[7],"compute":[8],"through":[9],"connections":[10],"to":[11,23,29,48,80,98,126,144],"boolean":[12],"circuits,":[13],"but":[14],"existing":[15],"results":[16,173],"lack":[17],"exact":[18],"characterizations":[19],"are":[21,37,95,124],"sensitive":[22],"modeling":[24],"choices.":[25],"Padded":[26],"--":[28,39],"whose":[30],"input":[31],"filler":[32],"symbols":[33],"such":[34],"as":[35,41,109],"``...''":[36],"appended":[38],"emerge":[40],"a":[42,61],"useful":[43],"gadget":[44],"for":[45,55,175],"establishing":[46],"equivalences":[47,76],"circuit":[49],"classes":[50],"by":[51],"providing":[52],"polynomial":[53],"space":[54],"adaptive":[56],"parallel":[57],"computation.":[58],"However,":[59],"only":[60],"limited":[62],"set":[63],"of":[64,100,136],"padded":[65,93,120],"transformer":[66],"idealizations":[67],"has":[68],"been":[69],"studied,":[70],"leaving":[71],"open":[72],"how":[73],"robustly":[74],"these":[75],"hold":[77,174],"under":[78,90],"changes":[79],"attention":[81,181],"type,":[82],"model":[83,107],"width,":[84],"uniformity.":[86],"We":[87],"find":[88],"that,":[89],"practical":[91],"assumptions,":[92],"surprisingly":[96],"robust":[97],"all":[99,171],"these,":[101],"identify":[103],"numeric":[104],"precision":[105,163],"depth":[108],"the":[110],"main":[111],"factors":[112],"affecting":[113],"expressivity.":[114],"Concretely,":[115],"we":[116],"prove":[117],"that":[118],"polynomially":[119],"$\\text{L-uniform}$":[121],"constant-precision":[122,148],"equivalent":[125],"$\\text{L-uniform":[127,133],"AC}^0$,":[128],"while":[129],"growing-precision":[130,154],"ones":[131,155],"achieve":[132],"TC}^0$":[134],"regardless":[135],"width.":[137],"Furthermore,":[138],"looping":[139],"enables":[140],"sequential":[141],"processing":[142],"analogous":[143],"circuits:":[145],"$\\log^d":[146],"N$-looped":[147],"reach":[150,156],"$\\text{FO-uniform":[151,157],"AC}^d$,":[152],"TC}^d$.":[158],"Interestingly,":[159],"growing":[160],"width":[161],"or":[162],"beyond":[164],"logarithmic":[165],"does":[166],"not":[167],"increase":[168],"expressivity,":[169],"our":[172],"both":[176],"softmax":[177],"average":[179],"hard":[180],"transformers.":[182]},"counts_by_year":[],"updated_date":"2026-06-11T09:08:48.828518","created_date":"2026-06-02T00:00:00"}