{"id":"https://openalex.org/W7155565028","doi":"https://doi.org/10.48550/arxiv.2604.21106","title":"How Much Is One Recurrence Worth? Iso-Depth Scaling Laws for Looped Language Models","display_name":"How Much Is One Recurrence Worth? Iso-Depth Scaling Laws for Looped Language Models","publication_year":2026,"publication_date":"2026-04-22","ids":{"openalex":"https://openalex.org/W7155565028","doi":"https://doi.org/10.48550/arxiv.2604.21106"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2604.21106","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.21106","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.2604.21106","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5134530353","display_name":"Kristian Schwethelm","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Schwethelm, Kristian","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5006461848","display_name":"Daniel Rueckert","orcid":"https://orcid.org/0000-0002-5683-5889"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Rueckert, Daniel","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5134543847","display_name":"Georgios Kaissis","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Kaissis, Georgios","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":0.3458000123500824,"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":0.3458000123500824,"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":0.29840001463890076,"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/T13629","display_name":"Text Readability and Simplification","score":0.04019999876618385,"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/block","display_name":"Block (permutation group theory)","score":0.6855000257492065},{"id":"https://openalex.org/keywords/scaling","display_name":"Scaling","score":0.6179999709129333},{"id":"https://openalex.org/keywords/measure","display_name":"Measure (data warehouse)","score":0.5861999988555908},{"id":"https://openalex.org/keywords/simple","display_name":"Simple (philosophy)","score":0.5307000279426575},{"id":"https://openalex.org/keywords/exponent","display_name":"Exponent","score":0.525600016117096},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.4887000024318695},{"id":"https://openalex.org/keywords/scaling-law","display_name":"Scaling law","score":0.47839999198913574},{"id":"https://openalex.org/keywords/joint","display_name":"Joint (building)","score":0.436599999666214}],"concepts":[{"id":"https://openalex.org/C2777210771","wikidata":"https://www.wikidata.org/wiki/Q4927124","display_name":"Block (permutation group theory)","level":2,"score":0.6855000257492065},{"id":"https://openalex.org/C99844830","wikidata":"https://www.wikidata.org/wiki/Q102441924","display_name":"Scaling","level":2,"score":0.6179999709129333},{"id":"https://openalex.org/C2780009758","wikidata":"https://www.wikidata.org/wiki/Q6804172","display_name":"Measure (data warehouse)","level":2,"score":0.5861999988555908},{"id":"https://openalex.org/C2780586882","wikidata":"https://www.wikidata.org/wiki/Q7520643","display_name":"Simple (philosophy)","level":2,"score":0.5307000279426575},{"id":"https://openalex.org/C2780388253","wikidata":"https://www.wikidata.org/wiki/Q5421508","display_name":"Exponent","level":2,"score":0.525600016117096},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5084999799728394},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.4887000024318695},{"id":"https://openalex.org/C2988430800","wikidata":"https://www.wikidata.org/wiki/Q428971","display_name":"Scaling law","level":3,"score":0.47839999198913574},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.45980000495910645},{"id":"https://openalex.org/C18555067","wikidata":"https://www.wikidata.org/wiki/Q8375051","display_name":"Joint (building)","level":2,"score":0.436599999666214},{"id":"https://openalex.org/C48677424","wikidata":"https://www.wikidata.org/wiki/Q6888088","display_name":"Mode (computer interface)","level":2,"score":0.35089999437332153},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.3449000120162964},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.33820000290870667},{"id":"https://openalex.org/C2778755073","wikidata":"https://www.wikidata.org/wiki/Q10858537","display_name":"Scale (ratio)","level":2,"score":0.33379998803138733},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3172999918460846},{"id":"https://openalex.org/C2777686260","wikidata":"https://www.wikidata.org/wiki/Q144037","display_name":"Calculus (dental)","level":2,"score":0.31279999017715454},{"id":"https://openalex.org/C2993597952","wikidata":"https://www.wikidata.org/wiki/Q108123644","display_name":"Block model","level":2,"score":0.30959999561309814},{"id":"https://openalex.org/C149364088","wikidata":"https://www.wikidata.org/wiki/Q185917","display_name":"Translation (biology)","level":4,"score":0.3037000000476837},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.30309998989105225},{"id":"https://openalex.org/C28826006","wikidata":"https://www.wikidata.org/wiki/Q33521","display_name":"Applied mathematics","level":1,"score":0.29750001430511475},{"id":"https://openalex.org/C192209626","wikidata":"https://www.wikidata.org/wiki/Q190909","display_name":"Focus (optics)","level":2,"score":0.29089999198913574},{"id":"https://openalex.org/C111335779","wikidata":"https://www.wikidata.org/wiki/Q3454686","display_name":"Reduction (mathematics)","level":2,"score":0.2881999909877777},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.2766000032424927},{"id":"https://openalex.org/C18762648","wikidata":"https://www.wikidata.org/wiki/Q42213","display_name":"Work (physics)","level":2,"score":0.26170000433921814},{"id":"https://openalex.org/C2776502983","wikidata":"https://www.wikidata.org/wiki/Q690182","display_name":"Contrast (vision)","level":2,"score":0.26030001044273376}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2604.21106","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.21106","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.2604.21106","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.21106","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":{"We":[0,143],"measure":[1,53],"how":[2],"much":[3],"one":[4],"recurrence":[5,23],"is":[6,69,170],"worth":[7],"to":[8,74,86,163,184],"a":[9,38,54,65,79,87,121,129,139,150,186,199],"looped":[10,123],"(depth-recurrent)":[11],"transformer,":[12],"in":[13,33,71,101],"equivalent":[14,70],"unique":[15,76,105],"parameters.":[16],"From":[17],"an":[18],"iso-depth":[19],"pretraining":[20],"sweep":[21],"across":[22],"counts":[24],"$r":[25],"\\in":[26],"\\{1,":[27],"2,":[28],"4,":[29],"8\\}$":[30],"spanning":[31],"${\\sim}50\\times$":[32],"training":[34,115,136],"compute,":[35],"we":[36],"fit":[37],"joint":[39],"scaling":[40],"law":[41],"$L":[42],"=":[43],"E":[44],"+":[45,47,50],"A\\,(N_\\text{once}":[46],"r^\u03c6":[48],"N_\\text{rec})^{-\u03b1}":[49],"B\\,D^{-\u03b2}$":[51],"and":[52],"recurrence-equivalence":[55],"exponent":[56],"$\u03c6=":[57,98],"0.46$.":[58],"Intuitively,":[59],"$\u03c6$":[60,148,162,183],"tells":[61],"us":[62],"whether":[63],"looping":[64],"block":[66,89],"$r$":[67,75],"times":[68],"validation":[72,111,177,202],"loss":[73,112,178,203],"blocks":[77,106],"of":[78,138,147],"non-looped":[80,131,141],"model":[81,124],"(full":[82],"equivalence,":[83],"$\u03c6{=}1$)":[84],"or":[85],"single":[88],"run":[90],"repeatedly":[91],"with":[92,107,128],"no":[93],"capacity":[94,188],"gain":[95],"($\u03c6{=}0$).":[96],"Our":[97,190],"0.46$":[99],"sits":[100],"between,":[102],"so":[103],"replacing":[104],"shared":[108],"recurrences":[109],"increases":[110],"at":[113,119],"matched":[114],"compute.":[116],"For":[117],"example,":[118],"$r{=}4$":[120],"410M":[122],"performs":[125],"on":[126,153],"par":[127],"580M":[130],"model,":[132],"but":[133],"incurs":[134],"the":[135,145,167],"cost":[137],"1B":[140],"one.":[142],"demonstrate":[144],"utility":[146],"as":[149],"diagnostic":[151],"tool":[152],"two":[154],"case":[155],"studies:":[156],"commonly":[157],"used":[158],"truncated":[159],"backpropagation":[160],"lowers":[161],"$0.38$,":[164],"indicating":[165],"that":[166],"loop":[168,194],"mechanism":[169],"poorly":[171],"trained":[172],"under":[173],"truncation,":[174],"even":[175],"though":[176],"decreases.":[179],"Conversely,":[180],"hyperconnections":[181],"raise":[182],"$0.65$,":[185],"genuine":[187],"gain.":[189],"method":[191],"separates":[192],"true":[193],"improvements":[195],"from":[196],"training-side":[197],"gains,":[198],"distinction":[200],"raw":[201],"cannot":[204],"make.":[205]},"counts_by_year":[],"updated_date":"2026-07-28T07:46:37.118299","created_date":"2026-04-25T00:00:00"}