{"id":"https://openalex.org/W7223260751","doi":"https://doi.org/10.48550/arxiv.2610.31089","title":"Transformers with CoT Discover Heavy-Ball Acceleration Through Training","display_name":"Transformers with CoT Discover Heavy-Ball Acceleration Through Training","publication_year":2026,"publication_date":"2026-10-22","ids":{"openalex":"https://openalex.org/W7223260751","doi":"https://doi.org/10.48550/arxiv.2610.31089"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2610.31089","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.31089","pdf_url":"https://arxiv.org/pdf/2610.31089","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":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2610.31089","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5159327304","display_name":"Yuxuan Cai","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I889458895","display_name":"University of Hong Kong","ror":"https://ror.org/02zhqgq86","country_code":"HK","type":"education","lineage":["https://openalex.org/I889458895"]}],"countries":["HK"],"is_corresponding":false,"raw_author_name":"Cai, Yuxuan","raw_affiliation_strings":["School of Computing & Data Science , The University of Hong Kong ;"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"School of Computing & Data Science , The University of Hong Kong ;","institution_ids":["https://openalex.org/I889458895"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5159297761","display_name":"Yuan Cao","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I889458895","display_name":"University of Hong Kong","ror":"https://ror.org/02zhqgq86","country_code":"HK","type":"education","lineage":["https://openalex.org/I889458895"]}],"countries":["HK"],"is_corresponding":false,"raw_author_name":"Cao, Yuan","raw_affiliation_strings":["School of Computing & Data Science , The University of Hong Kong ;"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"School of Computing & Data Science , The University of Hong Kong ;","institution_ids":["https://openalex.org/I889458895"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I889458895"],"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/T11612","display_name":"Stochastic Gradient Optimization Techniques","score":0.29490000009536743,"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/T11612","display_name":"Stochastic Gradient Optimization Techniques","score":0.29490000009536743,"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/T10028","display_name":"Topic Modeling","score":0.23819999396800995,"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":0.08229999989271164,"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/chain-of-thought-reasoning","display_name":"chain-of-thought reasoning","score":0.9620000123977661},{"id":"https://openalex.org/keywords/transformer","display_name":"transformer models","score":0.5350000262260437},{"id":"https://openalex.org/keywords/algorithm-discovery","display_name":"algorithm discovery","score":0.5849999785423279},{"id":"https://openalex.org/keywords/in-context-learning","display_name":"in-context learning","score":0.6600000262260437},{"id":"https://openalex.org/keywords/gradient-descent","display_name":"gradient descent","score":0.5849999785423279},{"id":"https://openalex.org/keywords/linear-regression","display_name":"linear regression","score":0.6290000081062317},{"id":"https://openalex.org/keywords/optimization-acceleration","display_name":"optimization acceleration","score":0.5299999713897705}],"concepts":[{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.7440000176429749},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5960000157356262},{"id":"https://openalex.org/C153258448","wikidata":"https://www.wikidata.org/wiki/Q1199743","display_name":"Gradient descent","level":3,"score":0.5717999935150146},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5699999928474426},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.4632999897003174},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.3677000105381012},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.3262999951839447},{"id":"https://openalex.org/C117896860","wikidata":"https://www.wikidata.org/wiki/Q11376","display_name":"Acceleration","level":2,"score":0.32510000467300415},{"id":"https://openalex.org/C83546350","wikidata":"https://www.wikidata.org/wiki/Q1139051","display_name":"Regression","level":2,"score":0.3199000060558319},{"id":"https://openalex.org/C175154964","wikidata":"https://www.wikidata.org/wiki/Q380077","display_name":"Task analysis","level":3,"score":0.30160000920295715},{"id":"https://openalex.org/C136389625","wikidata":"https://www.wikidata.org/wiki/Q334384","display_name":"Supervised learning","level":3,"score":0.28630000352859497},{"id":"https://openalex.org/C20162079","wikidata":"https://www.wikidata.org/wiki/Q1151406","display_name":"Case-based reasoning","level":2,"score":0.2702000141143799},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.25999999046325684},{"id":"https://openalex.org/C3017489831","wikidata":"https://www.wikidata.org/wiki/Q2393193","display_name":"Running time","level":2,"score":0.258899986743927},{"id":"https://openalex.org/C127413603","wikidata":"https://www.wikidata.org/wiki/Q11023","display_name":"Engineering","level":0,"score":0.25209999084472656},{"id":"https://openalex.org/C48921125","wikidata":"https://www.wikidata.org/wiki/Q10861030","display_name":"Linear regression","level":2,"score":0.2515000104904175}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2610.31089","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.31089","pdf_url":"https://arxiv.org/pdf/2610.31089","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":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2610.31089","doi":"https://doi.org/10.48550/arxiv.2610.31089","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2610.31089","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:2610.31089","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.31089","pdf_url":"https://arxiv.org/pdf/2610.31089","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":null,"license_id":null,"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/W7223260751.pdf","grobid_xml":"https://content.openalex.org/works/W7223260751.grobid-xml"},"referenced_works_count":10,"referenced_works":["https://openalex.org/W1967524849","https://openalex.org/W3094502228","https://openalex.org/W3215529831","https://openalex.org/W4221161695","https://openalex.org/W4385292374","https://openalex.org/W4406121759","https://openalex.org/W4416613006","https://openalex.org/W7160599110","https://openalex.org/W7163160246","https://openalex.org/W7163167451"],"related_works":["https://openalex.org/W6910801705","https://openalex.org/W7103149299","https://openalex.org/W4403444422","https://openalex.org/W7221815507","https://openalex.org/W7162650298","https://openalex.org/W4403322019","https://openalex.org/W4416159251","https://openalex.org/W4399417123","https://openalex.org/W4407903727","https://openalex.org/W4392012692"],"abstract_inverted_index":{"Modern":[0],"transformers":[1,32],"often":[2],"solve":[3],"complex":[4],"tasks":[5],"by":[6,26,45],"generating":[7],"and":[8,36],"refining":[9],"intermediate":[10],"reasoning":[11,78,141],"steps.":[12],"To":[13],"develop":[14],"a":[15,22,46,94,134],"theoretical":[16,136],"understanding":[17],"of":[18],"chain-of-thought":[19],"(CoT)":[20],"reasoning,":[21],"notable":[23],"recent":[24],"study":[25,67],"Huang":[27],"et":[28],"al.":[29],"(2025)":[30],"viewed":[31],"as":[33],"algorithm":[34,123,153],"executors,":[35],"showed":[37],"that":[38],"when":[39],"transformers'":[40],"CoT":[41,77],"steps":[42],"are":[43],"supervised":[44],"gradient-descent":[47,95],"teacher,":[48,96],"they":[49],"can":[50,79,142],"learn":[51],"to":[52,108],"implement":[53,109],"gradient":[54],"descent":[55],"for":[56,155],"in-context":[57,70],"weight":[58,71],"prediction":[59,72],"in":[60,138],"linear":[61],"regression.":[62],"In":[63],"this":[64,121],"paper,":[65],"we":[66],"the":[68,84,101,106,110,117,152],"same":[69],"task":[73],"but":[74],"ask":[75],"whether":[76],"go":[80],"beyond":[81],"merely":[82,150],"imitating":[83],"teacher":[85],"algorithm.":[86],"We":[87],"show":[88],"that,":[89],"after":[90],"initial":[91],"supervision":[92,99,126],"from":[93,127],"further":[97],"trajectory-level":[98],"using":[100],"true":[102],"regression":[103],"weights":[104],"leads":[105],"model":[107],"Heavy-Ball":[111,129],"method,":[112],"thereby":[113],"achieving":[114],"acceleration":[115],"over":[116],"original":[118],"teacher.":[119,130],"Notably,":[120],"accelerated":[122],"emerges":[124],"without":[125],"any":[128],"Our":[131],"results":[132],"provide":[133],"concrete":[135],"example":[137],"which":[139],"transformer":[140],"discover":[143],"new":[144],"algorithms":[145],"through":[146],"training,":[147],"rather":[148],"than":[149],"imitate":[151],"used":[154],"supervision.":[156]},"counts_by_year":[],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2026-10-24T00:00:00"}