{"id":"https://openalex.org/W7153025130","doi":"https://doi.org/10.48550/arxiv.2604.07822","title":"Loop, Think, & Generalize: Implicit Reasoning in Recurrent-Depth Transformers","display_name":"Loop, Think, & Generalize: Implicit Reasoning in Recurrent-Depth Transformers","publication_year":2026,"publication_date":"2026-04-09","ids":{"openalex":"https://openalex.org/W7153025130","doi":"https://doi.org/10.48550/arxiv.2604.07822"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2604.07822","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.07822","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":"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.2604.07822","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5001004876","display_name":"Harsh Kohli","orcid":"https://orcid.org/0000-0003-1431-6025"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Kohli, Harsh","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100755351","display_name":"Srinivasan Parthasarathy","orcid":"https://orcid.org/0000-0002-6062-6449"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Parthasarathy, Srinivasan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5133335787","display_name":"Huan Sun","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Sun, Huan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5049844853","display_name":"Yuekun Yao","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yao, Yuekun","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/T12090","display_name":"Language and cultural evolution","score":0.25360000133514404,"subfield":{"id":"https://openalex.org/subfields/3316","display_name":"Cultural Studies"},"field":{"id":"https://openalex.org/fields/33","display_name":"Social Sciences"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}},"topics":[{"id":"https://openalex.org/T12090","display_name":"Language and cultural evolution","score":0.25360000133514404,"subfield":{"id":"https://openalex.org/subfields/3316","display_name":"Cultural Studies"},"field":{"id":"https://openalex.org/fields/33","display_name":"Social Sciences"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}},{"id":"https://openalex.org/T10181","display_name":"Natural Language Processing Techniques","score":0.14630000293254852,"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/T12808","display_name":"Ferroelectric and Negative Capacitance Devices","score":0.1111999973654747,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/generalization","display_name":"Generalization","score":0.730400025844574},{"id":"https://openalex.org/keywords/transformer","display_name":"Transformer","score":0.6410999894142151},{"id":"https://openalex.org/keywords/memorization","display_name":"Memorization","score":0.5098999738693237},{"id":"https://openalex.org/keywords/parametric-statistics","display_name":"Parametric statistics","score":0.40299999713897705},{"id":"https://openalex.org/keywords/computation","display_name":"Computation","score":0.3962000012397766},{"id":"https://openalex.org/keywords/implicit-knowledge","display_name":"Implicit knowledge","score":0.3199000060558319},{"id":"https://openalex.org/keywords/mental-model","display_name":"Mental model","score":0.3093999922275543}],"concepts":[{"id":"https://openalex.org/C177148314","wikidata":"https://www.wikidata.org/wiki/Q170084","display_name":"Generalization","level":2,"score":0.730400025844574},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6769000291824341},{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.6410999894142151},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5763000249862671},{"id":"https://openalex.org/C30038468","wikidata":"https://www.wikidata.org/wiki/Q4354775","display_name":"Memorization","level":2,"score":0.5098999738693237},{"id":"https://openalex.org/C117251300","wikidata":"https://www.wikidata.org/wiki/Q1849855","display_name":"Parametric statistics","level":2,"score":0.40299999713897705},{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.3962000012397766},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.36719998717308044},{"id":"https://openalex.org/C2986065213","wikidata":"https://www.wikidata.org/wiki/Q743861","display_name":"Implicit knowledge","level":2,"score":0.3199000060558319},{"id":"https://openalex.org/C2982912361","wikidata":"https://www.wikidata.org/wiki/Q1851867","display_name":"Mental model","level":2,"score":0.3093999922275543},{"id":"https://openalex.org/C161301231","wikidata":"https://www.wikidata.org/wiki/Q3478658","display_name":"Knowledge representation and reasoning","level":2,"score":0.29089999198913574},{"id":"https://openalex.org/C99844830","wikidata":"https://www.wikidata.org/wiki/Q102441924","display_name":"Scaling","level":2,"score":0.27950000762939453},{"id":"https://openalex.org/C143587482","wikidata":"https://www.wikidata.org/wiki/Q1543216","display_name":"Iterative and incremental development","level":2,"score":0.27309998869895935},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.2676999866962433},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.26330000162124634},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.26089999079704285},{"id":"https://openalex.org/C2777379011","wikidata":"https://www.wikidata.org/wiki/Q938545","display_name":"Implicit learning","level":3,"score":0.25360000133514404},{"id":"https://openalex.org/C2780767217","wikidata":"https://www.wikidata.org/wiki/Q5532421","display_name":"Generality","level":2,"score":0.25200000405311584}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2604.07822","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.07822","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":"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.2604.07822","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.07822","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":"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","display_name":"Quality Education","score":0.6303393840789795}],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"We":[0,66,188],"study":[1,54,190],"implicit":[2,36,74],"reasoning,":[3,38],"i.e.":[4,79,93],"the":[5,62,73],"ability":[6,143],"to":[7,31,103,105,153,158,216],"combine":[8],"knowledge":[9,25,34,81],"or":[10],"rules":[11],"within":[12],"a":[13,40,146,204],"single":[14],"forward":[15],"pass.":[16],"While":[17],"transformer-based":[18],"large":[19],"language":[20],"models":[21,114],"store":[22],"substantial":[23],"factual":[24],"and":[26,90,156,202,213],"rules,":[27],"they":[28],"often":[29],"fail":[30],"compose":[32],"this":[33,51,142],"for":[35,86],"multi-hop":[37],"suggesting":[39],"lack":[41],"of":[42],"compositional":[43,69],"generalization":[44,70,127,155,171,215],"over":[45,61],"their":[46],"parametric":[47],"knowledge.":[48],"To":[49],"address":[50],"limitation,":[52,206],"we":[53,118,139,168],"recurrent-depth":[55,129,200],"transformers,":[56,201],"which":[57],"enables":[58],"iterative":[59],"computation":[60],"same":[63],"transformer":[64],"layers.":[65],"investigate":[67],"two":[68],"challenges":[71],"under":[72],"reasoning":[75,97],"scenario:":[76],"systematic":[77,137,159],"generalization,":[78,138,160],"combining":[80],"that":[82,120,141,170],"is":[83],"never":[84],"used":[85],"compositions":[87,107],"during":[88],"training,":[89],"depth":[91,98,166,174],"extrapolation,":[92,167,195],"generalizing":[94],"from":[95,116,151],"limited":[96],"(e.g.":[99,108],"training":[100,173,192,199],"on":[101,198],"up":[102],"5-hop)":[104],"deeper":[106,186],"10-hop).":[109],"Through":[110],"controlled":[111],"studies":[112],"with":[113,125,182],"trained":[115],"scratch,":[117],"show":[119,169],"while":[121],"vanilla":[122],"transformers":[123,130],"struggle":[124],"both":[126],"challenges,":[128],"can":[131,175],"effectively":[132],"make":[133],"such":[134],"generalization.":[135],"For":[136,165],"find":[140],"emerges":[144],"through":[145],"three-stage":[147],"grokking":[148],"process,":[149],"transitioning":[150],"memorization":[152],"in-distribution":[154],"finally":[157],"supported":[161],"by":[162,178],"mechanistic":[163],"analysis.":[164],"beyond":[172],"be":[176],"unlocked":[177],"scaling":[179],"inference-time":[180],"recurrence,":[181],"more":[183],"iterations":[184],"enabling":[185],"reasoning.":[187],"further":[189],"how":[191],"strategies":[193],"affect":[194],"providing":[196],"guidance":[197],"identify":[203],"key":[205],"overthinking,":[207],"where":[208],"excessive":[209],"recurrence":[210],"degrades":[211],"predictions":[212],"limits":[214],"very":[217],"deep":[218],"compositions.":[219]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-04-11T00:00:00"}