{"id":"https://openalex.org/W4387323828","doi":"https://doi.org/10.48550/arxiv.2310.00726","title":"Improving Length-Generalization in Transformers via Task Hinting","display_name":"Improving Length-Generalization in Transformers via Task Hinting","publication_year":2023,"publication_date":"2023-10-01","ids":{"openalex":"https://openalex.org/W4387323828","doi":"https://doi.org/10.48550/arxiv.2310.00726"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2310.00726","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2310.00726","pdf_url":"https://arxiv.org/pdf/2310.00726","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,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2310.00726","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5056617357","display_name":"Pranjal Awasthi","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I1291425158","display_name":"Google (United States)","ror":"https://ror.org/00njsd438","country_code":"US","type":"company","lineage":["https://openalex.org/I1291425158","https://openalex.org/I4210128969"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Awasthi, Pranjal","raw_affiliation_strings":["Google Research"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Google Research","institution_ids":["https://openalex.org/I1291425158"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5078381067","display_name":"Anupam Gupta","orcid":"https://orcid.org/0000-0001-5579-3405","observed_orcids":["https://orcid.org/0000-0001-5579-3405"]},"institutions":[{"id":"https://openalex.org/I1291425158","display_name":"Google (United States)","ror":"https://ror.org/00njsd438","country_code":"US","type":"company","lineage":["https://openalex.org/I1291425158","https://openalex.org/I4210128969"]},{"id":"https://openalex.org/I74973139","display_name":"Carnegie Mellon University","ror":"https://ror.org/05x2bcf33","country_code":"US","type":"education","lineage":["https://openalex.org/I74973139"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Gupta, Anupam","raw_affiliation_strings":["Carnegie Mellon University and Google Research"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Carnegie Mellon University and Google Research","institution_ids":["https://openalex.org/I1291425158","https://openalex.org/I74973139"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":2,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"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":0.6923999786376953,"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":0.6923999786376953,"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/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.0957999974489212,"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/T10320","display_name":"Neural Networks and Applications","score":0.03779999911785126,"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/length-generalization","display_name":"length generalization","score":0.9869999885559082},{"id":"https://openalex.org/keywords/transformer","display_name":"transformer models","score":0.7789999842643738},{"id":"https://openalex.org/keywords/auxiliary-tasks","display_name":"auxiliary tasks","score":0.7059999704360962},{"id":"https://openalex.org/keywords/multi-task-learning","display_name":"multi-task learning","score":0.9259999990463257},{"id":"https://openalex.org/keywords/sorting","display_name":"sorting","score":0.7360000014305115},{"id":"https://openalex.org/keywords/arithmetic-reasoning","display_name":"arithmetic reasoning","score":0.699999988079071},{"id":"https://openalex.org/keywords/probing","display_name":"probing","score":0.5720000267028809},{"id":"https://openalex.org/keywords/model-visualization","display_name":"model visualization","score":0.4909999966621399}],"concepts":[{"id":"https://openalex.org/C177148314","wikidata":"https://www.wikidata.org/wiki/Q170084","display_name":"Generalization","level":2,"score":0.7837538719177246},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7112962007522583},{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.6583857536315918},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.5709460973739624},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.49828529357910156},{"id":"https://openalex.org/C75553542","wikidata":"https://www.wikidata.org/wiki/Q178161","display_name":"A priori and a posteriori","level":2,"score":0.47117334604263306},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.43932071328163147},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.3380357623100281},{"id":"https://openalex.org/C94375191","wikidata":"https://www.wikidata.org/wiki/Q11205","display_name":"Arithmetic","level":1,"score":0.3324407935142517},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.17727655172348022},{"id":"https://openalex.org/C138885662","wikidata":"https://www.wikidata.org/wiki/Q5891","display_name":"Philosophy","level":0,"score":0.0},{"id":"https://openalex.org/C111472728","wikidata":"https://www.wikidata.org/wiki/Q9471","display_name":"Epistemology","level":1,"score":0.0},{"id":"https://openalex.org/C134306372","wikidata":"https://www.wikidata.org/wiki/Q7754","display_name":"Mathematical analysis","level":1,"score":0.0},{"id":"https://openalex.org/C162324750","wikidata":"https://www.wikidata.org/wiki/Q8134","display_name":"Economics","level":0,"score":0.0},{"id":"https://openalex.org/C62520636","wikidata":"https://www.wikidata.org/wiki/Q944","display_name":"Quantum mechanics","level":1,"score":0.0},{"id":"https://openalex.org/C165801399","wikidata":"https://www.wikidata.org/wiki/Q25428","display_name":"Voltage","level":2,"score":0.0},{"id":"https://openalex.org/C187736073","wikidata":"https://www.wikidata.org/wiki/Q2920921","display_name":"Management","level":1,"score":0.0},{"id":"https://openalex.org/C121332964","wikidata":"https://www.wikidata.org/wiki/Q413","display_name":"Physics","level":0,"score":0.0}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2310.00726","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2310.00726","pdf_url":"https://arxiv.org/pdf/2310.00726","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,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2310.00726","doi":"https://doi.org/10.48550/arxiv.2310.00726","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2310.00726","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,"listed_in":[],"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"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2310.00726","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2310.00726","pdf_url":"https://arxiv.org/pdf/2310.00726","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,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4387323828.pdf","grobid_xml":"https://content.openalex.org/works/W4387323828.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W7005554804","https://openalex.org/W4403788968","https://openalex.org/W4417308017","https://openalex.org/W4415914879","https://openalex.org/W4404088963","https://openalex.org/W4403882418","https://openalex.org/W4387801471","https://openalex.org/W4320087317","https://openalex.org/W4391871437","https://openalex.org/W7196931879"],"abstract_inverted_index":{"It":[0],"has":[1],"been":[2],"observed":[3,228],"in":[4,195,283],"recent":[5],"years":[6],"that":[7,72,120,131,183,240,277],"transformers":[8],"have":[9],"problems":[10],"with":[11,226],"length":[12,39,66,125,144,156,179,197,246],"generalization":[13,198],"for":[14],"certain":[15,38],"types":[16],"of":[17,26,51,141,155,178,231,245,267],"reasoning":[18],"and":[19,118,148,205,220],"arithmetic":[20],"tasks.":[21],"In":[22],"particular,":[23],"the":[24,52,75,86,100,150,210,215,218,227,232,250,256,265],"performance":[25,257],"a":[27,37,90,105,114,191,222,242],"transformer":[28],"model":[29,76,87,216],"trained":[30],"on":[31,61,77,138,153,235,258],"tasks":[32,187],"(say":[33],"addition)":[34],"up":[35],"to":[36,48,83,88,108,135,165,208],"(e.g.,":[40],"5":[41],"digit":[42],"numbers)":[43],"drops":[44],"sharply":[45],"when":[46],"applied":[47],"longer":[49],"instances":[50],"same":[53],"problem.":[54],"This":[55],"work":[56],"proposes":[57],"an":[58],"approach":[59,272],"based":[60,271],"task":[62,95,121,170,269],"hinting":[63,122,270],"towards":[64],"addressing":[65],"generalization.":[67,126,180],"Our":[68,172],"key":[69],"idea":[70],"is":[71,81,133],"while":[73,184],"training":[74,116,251],"task-specific":[78],"data,":[79],"it":[80,132],"helpful":[82],"simultaneously":[84],"train":[85,136],"solve":[89],"simpler":[91],"but":[92],"related":[93],"auxiliary":[94,186],"as":[96,104],"well.":[97],"We":[98,112,181,201],"study":[99,173],"classical":[101],"sorting":[102,128],"problem":[103],"canonical":[106],"example":[107],"evaluate":[109],"our":[110,236,268],"approach.":[111],"design":[113],"multitask":[115],"framework":[117],"show":[119,130,239,264],"significantly":[123],"improve":[124,149],"For":[127],"we":[129,238,262],"possible":[134],"models":[137],"data":[139],"consisting":[140],"sequences":[142,154],"having":[143],"at":[145],"most":[146],"$20$,":[147],"test":[151],"accuracy":[152],"$100$":[157],"from":[158],"less":[159],"than":[160,167],"1%":[161],"(for":[162],"standard":[163],"training)":[164],"more":[166],"92%":[168],"(via":[169],"hinting).":[171],"uncovers":[174],"several":[175,185],"interesting":[176],"aspects":[177],"observe":[182],"may":[188],"seem":[189],"natural":[190],"priori,":[192],"their":[193],"effectiveness":[194],"improving":[196],"differs":[199],"dramatically.":[200],"further":[202,254],"use":[203],"probing":[204],"visualization-based":[206],"techniques":[207,279],"understand":[209],"internal":[211],"mechanisms":[212],"via":[213],"which":[214],"performs":[217],"task,":[219],"propose":[221],"theoretical":[223],"construction":[224],"consistent":[225],"learning":[229],"behaviors":[230],"model.":[233],"Based":[234],"construction,":[237],"introducing":[241],"small":[243],"number":[244],"dependent":[247],"parameters":[248],"into":[249],"procedure":[252],"can":[253],"boost":[255],"unseen":[259],"lengths.":[260],"Finally,":[261],"also":[263],"efficacy":[266],"beyond":[273],"sorting,":[274],"giving":[275],"hope":[276],"these":[278],"will":[280],"be":[281],"applicable":[282],"broader":[284],"contexts.":[285]},"counts_by_year":[],"updated_date":"2026-09-30T09:27:27.532063","created_date":"2023-10-04T00:00:00"}