{"id":"https://openalex.org/W7220680654","doi":"https://doi.org/10.48550/arxiv.2610.06701","title":"A Solvable Model of Adaptive Learning Rate Rescaling: Acceleration, Stability & Scaling","display_name":"A Solvable Model of Adaptive Learning Rate Rescaling: Acceleration, Stability & Scaling","publication_year":2026,"publication_date":"2026-10-05","ids":{"openalex":"https://openalex.org/W7220680654","doi":"https://doi.org/10.48550/arxiv.2610.06701"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2610.06701","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.06701","pdf_url":"https://arxiv.org/pdf/2610.06701","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/2610.06701","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5138465384","display_name":"Itay Lavie","orcid":"https://orcid.org/0000-0003-4454-6454","observed_orcids":["https://orcid.org/0000-0003-4454-6454"]},"institutions":[{"id":"https://openalex.org/I136199984","display_name":"Harvard University","ror":"https://ror.org/03vek6s52","country_code":"US","type":"education","lineage":["https://openalex.org/I136199984"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Lavie, Itay","raw_affiliation_strings":["School of Engineering and Applied Sciences , Harvard University , Cambridge , MA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"School of Engineering and Applied Sciences , Harvard University , Cambridge , MA","institution_ids":["https://openalex.org/I136199984"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5086922438","display_name":"Clarissa Lauditi","orcid":"https://orcid.org/0000-0001-8472-7562","observed_orcids":["https://orcid.org/0000-0001-8472-7562"]},"institutions":[{"id":"https://openalex.org/I136199984","display_name":"Harvard University","ror":"https://ror.org/03vek6s52","country_code":"US","type":"education","lineage":["https://openalex.org/I136199984"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Lauditi, Clarissa","raw_affiliation_strings":["School of Engineering and Applied Sciences , Harvard University , Cambridge , MA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"School of Engineering and Applied Sciences , Harvard University , Cambridge , MA","institution_ids":["https://openalex.org/I136199984"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5023195984","display_name":"Cengiz Pehlevan","orcid":"https://orcid.org/0000-0001-9767-6063","observed_orcids":["https://orcid.org/0000-0001-9767-6063"]},"institutions":[{"id":"https://openalex.org/I136199984","display_name":"Harvard University","ror":"https://ror.org/03vek6s52","country_code":"US","type":"education","lineage":["https://openalex.org/I136199984"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Pehlevan, Cengiz","raw_affiliation_strings":["Center for Brain Science , Harvard University , Cambridge , MA","Kempner Institute for the Study of Natural and Artificial Intelligence , Harvard University , Cambridge , MA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Center for Brain Science , Harvard University , Cambridge , MA","institution_ids":["https://openalex.org/I136199984"]},{"raw_affiliation_string":"Kempner Institute for the Study of Natural and Artificial Intelligence , Harvard University , Cambridge , MA","institution_ids":["https://openalex.org/I136199984"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I136199984"],"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":0.9994999766349792,"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":0.9994999766349792,"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":0.00019999999494757503,"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/T12535","display_name":"Machine Learning and Data Classification","score":9.999999747378752e-05,"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/gradient-normalization","display_name":"gradient normalization","score":0.49900001287460327},{"id":"https://openalex.org/keywords/random-feature-models","display_name":"random feature models","score":0.7300000190734863},{"id":"https://openalex.org/keywords/dynamical-mean-field-theory","display_name":"dynamical mean-field theory","score":0.9150000214576721},{"id":"https://openalex.org/keywords/edge-of-stability","display_name":"edge of stability","score":0.6190000176429749},{"id":"https://openalex.org/keywords/optimization-acceleration","display_name":"optimization acceleration","score":0.6230000257492065},{"id":"https://openalex.org/keywords/training-stability","display_name":"training stability","score":0.5490000247955322},{"id":"https://openalex.org/keywords/batch-size-scaling","display_name":"batch size scaling","score":0.8080000281333923},{"id":"https://openalex.org/keywords/resource-scaling","display_name":"resource scaling","score":0.8190000057220459},{"id":"https://openalex.org/keywords/finite-time-convergence","display_name":"finite-time convergence","score":0.6639999747276306}],"concepts":[{"id":"https://openalex.org/C99844830","wikidata":"https://www.wikidata.org/wiki/Q102441924","display_name":"Scaling","level":2,"score":0.6933000087738037},{"id":"https://openalex.org/C136886441","wikidata":"https://www.wikidata.org/wiki/Q926129","display_name":"Normalization (sociology)","level":2,"score":0.6657999753952026},{"id":"https://openalex.org/C75235859","wikidata":"https://www.wikidata.org/wiki/Q582659","display_name":"Exponential growth","level":2,"score":0.5925999879837036},{"id":"https://openalex.org/C2777303404","wikidata":"https://www.wikidata.org/wiki/Q759757","display_name":"Convergence (economics)","level":2,"score":0.5688999891281128},{"id":"https://openalex.org/C112972136","wikidata":"https://www.wikidata.org/wiki/Q7595718","display_name":"Stability (learning theory)","level":2,"score":0.5115000009536743},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.5112000107765198},{"id":"https://openalex.org/C28826006","wikidata":"https://www.wikidata.org/wiki/Q33521","display_name":"Applied mathematics","level":1,"score":0.4934000074863434},{"id":"https://openalex.org/C186633575","wikidata":"https://www.wikidata.org/wiki/Q845060","display_name":"Maxima and minima","level":2,"score":0.4860999882221222},{"id":"https://openalex.org/C50488108","wikidata":"https://www.wikidata.org/wiki/Q618684","display_name":"Marginal stability","level":3,"score":0.4706999957561493},{"id":"https://openalex.org/C117896860","wikidata":"https://www.wikidata.org/wiki/Q11376","display_name":"Acceleration","level":2,"score":0.4641000032424927},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.4325000047683716},{"id":"https://openalex.org/C151376022","wikidata":"https://www.wikidata.org/wiki/Q168698","display_name":"Exponential function","level":2,"score":0.43209999799728394},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.4041999876499176},{"id":"https://openalex.org/C2780388253","wikidata":"https://www.wikidata.org/wiki/Q5421508","display_name":"Exponent","level":2,"score":0.4009000062942505},{"id":"https://openalex.org/C57869625","wikidata":"https://www.wikidata.org/wiki/Q1783502","display_name":"Rate of convergence","level":3,"score":0.3880999982357025},{"id":"https://openalex.org/C774472","wikidata":"https://www.wikidata.org/wiki/Q6760393","display_name":"Margin (machine learning)","level":2,"score":0.34529998898506165},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.3203999996185303},{"id":"https://openalex.org/C147764199","wikidata":"https://www.wikidata.org/wiki/Q6865248","display_name":"Minification","level":2,"score":0.31049999594688416},{"id":"https://openalex.org/C132459708","wikidata":"https://www.wikidata.org/wiki/Q744069","display_name":"Extrapolation","level":2,"score":0.302700012922287},{"id":"https://openalex.org/C47446073","wikidata":"https://www.wikidata.org/wiki/Q5165890","display_name":"Control theory (sociology)","level":3,"score":0.2946999967098236},{"id":"https://openalex.org/C88548561","wikidata":"https://www.wikidata.org/wiki/Q347599","display_name":"sort","level":2,"score":0.290800005197525},{"id":"https://openalex.org/C2776359362","wikidata":"https://www.wikidata.org/wiki/Q2145286","display_name":"Representation (politics)","level":3,"score":0.28940001130104065},{"id":"https://openalex.org/C115680565","wikidata":"https://www.wikidata.org/wiki/Q5977448","display_name":"Gradient method","level":2,"score":0.2757999897003174},{"id":"https://openalex.org/C77553402","wikidata":"https://www.wikidata.org/wiki/Q13222579","display_name":"Upper and lower bounds","level":2,"score":0.26109999418258667},{"id":"https://openalex.org/C91528185","wikidata":"https://www.wikidata.org/wiki/Q520952","display_name":"Maxima","level":3,"score":0.2565000057220459},{"id":"https://openalex.org/C2778755073","wikidata":"https://www.wikidata.org/wiki/Q10858537","display_name":"Scale (ratio)","level":2,"score":0.2535000145435333}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2610.06701","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.06701","pdf_url":"https://arxiv.org/pdf/2610.06701","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.2610.06701","doi":"https://doi.org/10.48550/arxiv.2610.06701","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2610.06701","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:2610.06701","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.06701","pdf_url":"https://arxiv.org/pdf/2610.06701","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":[{"id":"https://openalex.org/G381467788","display_name":"CAREER: Developing Neural Network Theory for Uncovering How the Brain Learns","funder_award_id":"2239780","funder_id":"https://openalex.org/F4320306076","funder_display_name":"National Science Foundation"}],"funders":[{"id":"https://openalex.org/F4320306076","display_name":"National Science Foundation","ror":"https://ror.org/021nxhr62"},{"id":"https://openalex.org/F4320332180","display_name":"Defense Advanced Research Projects Agency","ror":"https://ror.org/02caytj08"}],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7220680654.pdf","grobid_xml":"https://content.openalex.org/works/W7220680654.grobid-xml"},"referenced_works_count":34,"referenced_works":["https://openalex.org/W2766164908","https://openalex.org/W2775461895","https://openalex.org/W2900167092","https://openalex.org/W2903697572","https://openalex.org/W2908510526","https://openalex.org/W2964121744","https://openalex.org/W2996642285","https://openalex.org/W3001279689","https://openalex.org/W3008829306","https://openalex.org/W3129910129","https://openalex.org/W3132026801","https://openalex.org/W3133450183","https://openalex.org/W4281251260","https://openalex.org/W4289446290","https://openalex.org/W4307929880","https://openalex.org/W4380715436","https://openalex.org/W4391555764","https://openalex.org/W4394654497","https://openalex.org/W4396600484","https://openalex.org/W4403707245","https://openalex.org/W4403815243","https://openalex.org/W4405956499","https://openalex.org/W4412955598","https://openalex.org/W4414595198","https://openalex.org/W4414846785","https://openalex.org/W7092285716","https://openalex.org/W7127646912","https://openalex.org/W7128504444","https://openalex.org/W7133190866","https://openalex.org/W7133225605","https://openalex.org/W7160834952","https://openalex.org/W7160952933","https://openalex.org/W7163508073","https://openalex.org/W7171459003"],"related_works":["https://openalex.org/W7212076696","https://openalex.org/W7127923911","https://openalex.org/W2944082518","https://openalex.org/W2892501656","https://openalex.org/W3004988528","https://openalex.org/W7139099463","https://openalex.org/W3091985885","https://openalex.org/W4307926947","https://openalex.org/W3034228994","https://openalex.org/W4416048536"],"abstract_inverted_index":{"A":[0],"recurring":[1],"design":[2],"principle":[3],"in":[4,35,153],"modern":[5],"optimizers":[6],"is":[7],"to":[8,89,120,175],"decouple":[9],"update":[10],"magnitude":[11],"from":[12],"the":[13,65,69,84,111,116,165,184],"raw":[14],"gradient":[15],"norm,":[16],"yet":[17],"its":[18],"consequences":[19],"for":[20,103],"learning-curve":[21],"and":[22,42,76,99,118,130,188,199],"resource":[23],"scaling":[24],"remain":[25],"unclear.":[26],"We":[27,57,151],"isolate":[28],"this":[29],"mechanism":[30],"by":[31,171],"studying":[32],"normalized":[33],"SGD":[34],"a":[36,59,203],"random-feature":[37],"model":[38,74],"with":[39,93],"power-law":[40,85],"teacher":[41],"data":[43],"covariance.":[44],"Fixed-norm":[45],"updates":[46],"induce":[47],"an":[48],"effective":[49],"learning":[50],"rate":[51],"that":[52],"grows":[53],"as":[54],"gradients":[55],"shrink.":[56],"derive":[58],"dynamical":[60],"mean-field":[61],"theory":[62,125],"(DMFT)":[63],"describing":[64],"joint":[66],"dependence":[67],"of":[68,155],"loss":[70],"on":[71,181],"training":[72,146],"time,":[73],"width":[75,162],"batch":[77,159],"size.":[78],"Normalization":[79],"initially":[80],"accelerates":[81],"SGD,":[82],"mapping":[83],"exponent":[86],"$r_{\\rm":[87,97,104],"SGD}<1$":[88],"$2r_{\\rm":[90],"SGD}/(1-r_{\\rm":[91],"SGD})$,":[92],"exponential":[94],"convergence":[95,102],"at":[96,148],"SGD}=1$":[98],"formal":[100],"finite-time":[101],"SGD}>1$.":[105],"At":[106],"finite":[107],"step":[108,170],"size,":[109],"however,":[110],"same":[112],"feedback":[113],"ultimately":[114],"breaks":[115],"acceleration":[117],"leads":[119],"marginal":[121],"stability.":[122],"The":[123],"late-time":[124],"yields":[126],"width-limited,":[127],"edge-of-stochastic-stability":[128],"(EoSS),":[129],"deterministic":[131],"edge-of-stability":[132],"(EoS)":[133],"regimes.":[134],"These":[135],"phases":[136,157],"determine":[137],"when":[138],"larger":[139],"batches":[140],"or":[141,161],"wider":[142],"models":[143],"reduce":[144],"serial":[145],"time":[147],"comparable":[149],"compute.":[150],"quantify":[152],"which":[154],"these":[156,192],"increased":[158],"size":[160],"can":[163],"compensate":[164],"excess":[166],"compute":[167],"use":[168],"per":[169],"fewer":[172],"optimization":[173],"steps":[174],"target":[176],"loss.":[177],"Linearized":[178],"ResNet":[179],"experiments":[180],"CIFAR-5M":[182],"support":[183],"predicted":[185],"acceleration,":[186,196],"breakdown,":[187],"resource-scaling":[189],"trends.":[190],"Together,":[191],"results":[193],"connect":[194],"normalization-induced":[195],"EoS":[197],"effects,":[198],"width--batch":[200],"allocation":[201],"within":[202],"solvable":[204],"theory.":[205]},"counts_by_year":[],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2026-10-07T00:00:00"}