{"id":"https://openalex.org/W7160645316","doi":"https://doi.org/10.48550/arxiv.2605.05794","title":"Revealing Modular Gradient Noise Imbalance in LLMs: Calibrating Adam via Signal-to-Noise Ratio","display_name":"Revealing Modular Gradient Noise Imbalance in LLMs: Calibrating Adam via Signal-to-Noise Ratio","publication_year":2026,"publication_date":"2026-05-07","ids":{"openalex":"https://openalex.org/W7160645316","doi":"https://doi.org/10.48550/arxiv.2605.05794"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.05794","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.05794","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.2605.05794","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5135670510","display_name":"Ziqing Wen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wen, Ziqing","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5115595215","display_name":"Zhouyang Liu","orcid":"https://orcid.org/0009-0007-6424-2622"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liu, Zhouyang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5073951717","display_name":"Jiahuan Wang","orcid":"https://orcid.org/0000-0002-2173-8608"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Jiahuan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135644442","display_name":"Ping Luo","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Luo, Ping","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135697881","display_name":"Li Shen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Shen, Li","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135664408","display_name":"Dongsheng Li","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Dongsheng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5135693128","display_name":"Tao Sun","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Sun, Tao","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":0.22509999573230743,"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":0.22509999573230743,"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/T11714","display_name":"Multimodal Machine Learning Applications","score":0.06840000301599503,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"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/T11948","display_name":"Machine Learning in Materials Science","score":0.06790000200271606,"subfield":{"id":"https://openalex.org/subfields/2505","display_name":"Materials Chemistry"},"field":{"id":"https://openalex.org/fields/25","display_name":"Materials Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/modular-design","display_name":"Modular design","score":0.6503000259399414},{"id":"https://openalex.org/keywords/convergence","display_name":"Convergence (economics)","score":0.5317999720573425},{"id":"https://openalex.org/keywords/scaling","display_name":"Scaling","score":0.5311999917030334},{"id":"https://openalex.org/keywords/noise","display_name":"Noise (video)","score":0.4675999879837036},{"id":"https://openalex.org/keywords/rate-of-convergence","display_name":"Rate of convergence","score":0.4652000069618225},{"id":"https://openalex.org/keywords/scale","display_name":"Scale (ratio)","score":0.44440001249313354},{"id":"https://openalex.org/keywords/key","display_name":"Key (lock)","score":0.31439998745918274}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7020000219345093},{"id":"https://openalex.org/C101468663","wikidata":"https://www.wikidata.org/wiki/Q1620158","display_name":"Modular design","level":2,"score":0.6503000259399414},{"id":"https://openalex.org/C2777303404","wikidata":"https://www.wikidata.org/wiki/Q759757","display_name":"Convergence (economics)","level":2,"score":0.5317999720573425},{"id":"https://openalex.org/C99844830","wikidata":"https://www.wikidata.org/wiki/Q102441924","display_name":"Scaling","level":2,"score":0.5311999917030334},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4778999984264374},{"id":"https://openalex.org/C99498987","wikidata":"https://www.wikidata.org/wiki/Q2210247","display_name":"Noise (video)","level":3,"score":0.4675999879837036},{"id":"https://openalex.org/C57869625","wikidata":"https://www.wikidata.org/wiki/Q1783502","display_name":"Rate of convergence","level":3,"score":0.4652000069618225},{"id":"https://openalex.org/C2778755073","wikidata":"https://www.wikidata.org/wiki/Q10858537","display_name":"Scale (ratio)","level":2,"score":0.44440001249313354},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.3702999949455261},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.3321000039577484},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.31439998745918274},{"id":"https://openalex.org/C115680565","wikidata":"https://www.wikidata.org/wiki/Q5977448","display_name":"Gradient method","level":2,"score":0.30979999899864197},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.30730000138282776},{"id":"https://openalex.org/C137836250","wikidata":"https://www.wikidata.org/wiki/Q984063","display_name":"Optimization problem","level":2,"score":0.27799999713897705},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.2761000096797943},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.26980000734329224},{"id":"https://openalex.org/C2987595161","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Optimization algorithm","level":2,"score":0.2669000029563904},{"id":"https://openalex.org/C34559072","wikidata":"https://www.wikidata.org/wiki/Q2334061","display_name":"Design of experiments","level":2,"score":0.25609999895095825}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.05794","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.05794","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.2605.05794","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.05794","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":{"The":[0],"impressive":[1],"performance":[2,137],"of":[3,90],"large":[4],"language":[5],"models":[6],"(LLMs)":[7],"arises":[8],"from":[9],"their":[10],"massive":[11],"scale":[12,109],"and":[13,70,95,134],"heterogeneous":[14],"module":[15],"composition.":[16],"However,":[17],"this":[18],"structural":[19],"heterogeneity":[20],"introduces":[21],"additional":[22],"optimization":[23,64],"challenges.":[24],"While":[25],"adaptive":[26],"optimizers":[27],"such":[28],"as":[29],"Adam(W)":[30],"provide":[31],"per-parameter":[32],"adaptivity,":[33],"they":[34],"do":[35],"not":[36],"explicitly":[37],"account":[38],"for":[39],"module-level":[40,106],"gradient":[41],"heterogeneity,":[42],"resulting":[43],"in":[44,92],"slower":[45],"convergence,":[46],"suboptimal":[47],"performance,":[48],"or":[49,62,76],"training":[50,126,150],"instability.":[51],"Existing":[52],"approaches":[53],"typically":[54],"rely":[55],"on":[56],"manually":[57],"tuned":[58,141],"module-specific":[59,142],"learning":[60,115,143],"rates":[61],"specific":[63],"strategies,":[65],"which":[66],"are":[67],"computationally":[68],"costly":[69],"difficult":[71],"to":[72,108,139],"generalize":[73],"across":[74],"tasks":[75],"models.":[77],"To":[78],"establish":[79],"a":[80],"more":[81],"principled":[82],"approach,":[83],"we":[84],"first":[85],"analyze":[86],"the":[87],"noise-damping":[88],"behavior":[89],"Adam":[91,110],"high-noise":[93],"modules":[94],"introduce":[96],"\\textbf{Module-wise":[97],"Learning":[98],"Rate":[99],"Scaling":[100],"via":[101],"SNR":[102],"(MoLS)}.":[103],"MoLS":[104,130],"estimates":[105],"SNRs":[107],"updates,":[111],"allowing":[112],"automated":[113],"module-wise":[114],"rate":[116],"allocation":[117],"without":[118],"manual":[119],"tuning.":[120],"Empirical":[121],"results":[122],"through":[123],"multiple":[124],"LLM":[125],"benchmarks":[127],"demonstrate":[128],"that":[129],"improves":[131],"convergence":[132],"speed":[133],"generalization,":[135],"achieving":[136],"comparable":[138],"carefully":[140],"rates,":[144],"while":[145],"remaining":[146],"compatible":[147],"with":[148],"memory-efficient":[149],"algorithms.":[151]},"counts_by_year":[],"updated_date":"2026-07-28T07:46:37.118299","created_date":"2026-05-09T00:00:00"}