{"id":"https://openalex.org/W7204558753","doi":"https://doi.org/10.48550/arxiv.2608.26973","title":"Squeezing More from Limited Data with Recursive Transformers","display_name":"Squeezing More from Limited Data with Recursive Transformers","publication_year":2026,"publication_date":"2026-08-27","ids":{"openalex":"https://openalex.org/W7204558753","doi":"https://doi.org/10.48550/arxiv.2608.26973"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2608.26973","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.26973","pdf_url":"https://arxiv.org/pdf/2608.26973","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/2608.26973","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5148537326","display_name":"Serdar G\u00fclbahar","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I62916508","display_name":"Technical University of Munich","ror":"https://ror.org/02kkvpp62","country_code":"DE","type":"education","lineage":["https://openalex.org/I62916508"]}],"countries":["DE"],"is_corresponding":false,"raw_author_name":"G\u00fclbahar, Serdar","raw_affiliation_strings":["Technical University of Munich"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Technical University of Munich","institution_ids":["https://openalex.org/I62916508"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5114452403","display_name":"Lukas Edman","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Edman, Lukas","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5137194995","display_name":"Alexander Fraser","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Fraser, Alexander","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"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":0.9423999786376953,"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":0.9423999786376953,"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":0.039799999445676804,"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":0.008500000461935997,"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/model-scaling","display_name":"model scaling","score":0.5379999876022339},{"id":"https://openalex.org/keywords/overfitting","display_name":"overfitting","score":0.765999972820282},{"id":"https://openalex.org/keywords/vocabulary-mapping","display_name":"vocabulary mapping","score":0.4970000088214874},{"id":"https://openalex.org/keywords/parameter-efficiency","display_name":"parameter efficiency","score":0.6499999761581421},{"id":"https://openalex.org/keywords/language-model-pretraining","display_name":"language model pretraining","score":0.453000009059906}],"concepts":[{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.5950999855995178},{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.5774000287055969},{"id":"https://openalex.org/C22019652","wikidata":"https://www.wikidata.org/wiki/Q331309","display_name":"Overfitting","level":3,"score":0.5685999989509583},{"id":"https://openalex.org/C99844830","wikidata":"https://www.wikidata.org/wiki/Q102441924","display_name":"Scaling","level":2,"score":0.5641000270843506},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5602999925613403},{"id":"https://openalex.org/C177148314","wikidata":"https://www.wikidata.org/wiki/Q170084","display_name":"Generalization","level":2,"score":0.4903999865055084},{"id":"https://openalex.org/C2776207758","wikidata":"https://www.wikidata.org/wiki/Q5303302","display_name":"Downstream (manufacturing)","level":2,"score":0.4487000107765198},{"id":"https://openalex.org/C2777210771","wikidata":"https://www.wikidata.org/wiki/Q4927124","display_name":"Block (permutation group theory)","level":2,"score":0.4131999909877777},{"id":"https://openalex.org/C206588197","wikidata":"https://www.wikidata.org/wiki/Q846574","display_name":"Reuse","level":2,"score":0.41110000014305115},{"id":"https://openalex.org/C2778755073","wikidata":"https://www.wikidata.org/wiki/Q10858537","display_name":"Scale (ratio)","level":2,"score":0.3693999946117401},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.3573000133037567},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.3497999906539917},{"id":"https://openalex.org/C206729178","wikidata":"https://www.wikidata.org/wiki/Q2271896","display_name":"Scheduling (production processes)","level":2,"score":0.3296999931335449},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.3057999908924103},{"id":"https://openalex.org/C149629883","wikidata":"https://www.wikidata.org/wiki/Q660926","display_name":"Fraction (chemistry)","level":2,"score":0.30239999294281006},{"id":"https://openalex.org/C4679612","wikidata":"https://www.wikidata.org/wiki/Q866298","display_name":"Aggregate (composite)","level":2,"score":0.30140000581741333},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.29490000009536743}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2608.26973","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.26973","pdf_url":"https://arxiv.org/pdf/2608.26973","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.2608.26973","doi":"https://doi.org/10.48550/arxiv.2608.26973","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2608.26973","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:2608.26973","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.26973","pdf_url":"https://arxiv.org/pdf/2608.26973","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":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7204558753.pdf","grobid_xml":"https://content.openalex.org/works/W7204558753.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W7154162121","https://openalex.org/W4417082077","https://openalex.org/W4389519173","https://openalex.org/W3199241049","https://openalex.org/W7171934181","https://openalex.org/W4415318403","https://openalex.org/W4407123799","https://openalex.org/W4404314310","https://openalex.org/W7163200956","https://openalex.org/W4396821442"],"abstract_inverted_index":{"Pre-training":[0],"under":[1],"limited":[2],"data":[3,17,66],"requires":[4],"a":[5,15,86,110],"different":[6],"view":[7],"of":[8,89],"scaling":[9],"than":[10],"web-scale":[11],"language":[12],"modeling.":[13],"With":[14],"fixed":[16],"budget":[18,67,92],"but":[19],"relatively":[20],"abundant":[21],"compute,":[22,117],"increasing":[23],"parameter":[24,91],"count":[25],"helps":[26],"only":[27],"up":[28],"to":[29,80,98,115,122],"an":[30],"optimal":[31,59],"scale;":[32],"beyond":[33],"that":[34,58,74,133],"point,":[35],"models":[36,130],"overfit":[37],"and":[38,52,56,68,93,131,140],"generalization":[39],"worsens.":[40],"We":[41,72,126],"study":[42,106],"this":[43,81,103],"behavior":[44],"across":[45,113],"10M-100M":[46],"word":[47],"pre-training":[48],"budgets,":[49],"two":[50],"corpora,":[51],"multiple":[53],"downstream":[54,70],"evaluations,":[55],"find":[57,132],"size":[60],"depends":[61],"strongly":[62],"on":[63],"both":[64],"the":[65,69,90],"target.":[71],"argue":[73],"standard":[75,136],"Transformers":[76,137],"scale":[77,116],"down":[78],"poorly":[79],"setting,":[82],"because":[83],"embeddings":[84,121],"consume":[85],"large":[87],"fraction":[88],"per-token":[94],"computation":[95],"is":[96],"tied":[97],"representational":[99],"capacity.":[100],"To":[101],"address":[102],"coupling,":[104],"we":[105],"recursive":[107,129],"Transformers,":[108],"reusing":[109],"shared":[111],"block":[112],"depth":[114],"together":[118],"with":[119,146],"factorized":[120],"reduce":[123],"vocabulary-map":[124],"parameters.":[125],"train":[127],"three":[128],"they":[134],"outperform":[135],"at":[138],"10M":[139],"100M":[141],"words,":[142],"while":[143],"remaining":[144],"competitive":[145],"BabyLM":[147],"Challenge":[148],"2025":[149],"winners.":[150]},"counts_by_year":[],"updated_date":"2026-10-28T10:33:24.094632","created_date":"2026-08-29T00:00:00"}