{"id":"https://openalex.org/W7160302161","doi":"https://doi.org/10.48550/arxiv.2605.02853","title":"Trust, but Verify: Peeling Low-Bit Transformer Networks for Training Monitoring","display_name":"Trust, but Verify: Peeling Low-Bit Transformer Networks for Training Monitoring","publication_year":2026,"publication_date":"2026-05-04","ids":{"openalex":"https://openalex.org/W7160302161","doi":"https://doi.org/10.48550/arxiv.2605.02853"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.02853","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.02853","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":"article"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2605.02853","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5021150988","display_name":"Arian Eamaz","orcid":"https://orcid.org/0000-0003-0479-9502"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Eamaz, Arian","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5067897446","display_name":"Farhang Yeganegi","orcid":"https://orcid.org/0000-0003-0742-7667"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yeganegi, Farhang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5053865178","display_name":"Mojtaba Soltanalian","orcid":"https://orcid.org/0000-0003-3714-4449"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Soltanalian, Mojtaba","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":3,"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/T10036","display_name":"Advanced Neural Network Applications","score":0.24889999628067017,"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"}},"topics":[{"id":"https://openalex.org/T10036","display_name":"Advanced Neural Network Applications","score":0.24889999628067017,"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/T12611","display_name":"Neural Networks and Reservoir Computing","score":0.12039999663829803,"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/T11948","display_name":"Machine Learning in Materials Science","score":0.09459999948740005,"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/transformer","display_name":"Transformer","score":0.7225000262260437},{"id":"https://openalex.org/keywords/training","display_name":"Training (meteorology)","score":0.5396999716758728},{"id":"https://openalex.org/keywords/training-set","display_name":"Training set","score":0.5067999958992004},{"id":"https://openalex.org/keywords/artificial-neural-network","display_name":"Artificial neural network","score":0.5047000050544739},{"id":"https://openalex.org/keywords/deep-neural-networks","display_name":"Deep neural networks","score":0.47839999198913574},{"id":"https://openalex.org/keywords/convergence","display_name":"Convergence (economics)","score":0.43070000410079956}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7289999723434448},{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.7225000262260437},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.5396999716758728},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.51910001039505},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.5067999958992004},{"id":"https://openalex.org/C50644808","wikidata":"https://www.wikidata.org/wiki/Q192776","display_name":"Artificial neural network","level":2,"score":0.5047000050544739},{"id":"https://openalex.org/C2984842247","wikidata":"https://www.wikidata.org/wiki/Q197536","display_name":"Deep neural networks","level":3,"score":0.47839999198913574},{"id":"https://openalex.org/C2777303404","wikidata":"https://www.wikidata.org/wiki/Q759757","display_name":"Convergence (economics)","level":2,"score":0.43070000410079956},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.39809998869895935},{"id":"https://openalex.org/C108583219","wikidata":"https://www.wikidata.org/wiki/Q197536","display_name":"Deep learning","level":2,"score":0.34619998931884766},{"id":"https://openalex.org/C2986605239","wikidata":"https://www.wikidata.org/wiki/Q925667","display_name":"Numerical models","level":3,"score":0.2831999957561493},{"id":"https://openalex.org/C59404180","wikidata":"https://www.wikidata.org/wiki/Q17013334","display_name":"Feature learning","level":2,"score":0.27720001339912415},{"id":"https://openalex.org/C67186912","wikidata":"https://www.wikidata.org/wiki/Q367664","display_name":"Data modeling","level":2,"score":0.26750001311302185},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.2660999894142151},{"id":"https://openalex.org/C4679612","wikidata":"https://www.wikidata.org/wiki/Q866298","display_name":"Aggregate (composite)","level":2,"score":0.2565000057220459},{"id":"https://openalex.org/C79403827","wikidata":"https://www.wikidata.org/wiki/Q3988","display_name":"Real-time computing","level":1,"score":0.2506999969482422}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.02853","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.02853","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":"article"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2605.02853","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.02853","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":"article"},"sustainable_development_goals":[{"score":0.500740110874176,"display_name":"Quality Education","id":"https://metadata.un.org/sdg/4"}],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Understanding":[0],"whether":[1],"deep":[2],"neural":[3],"networks":[4],"are":[5,41,156,177],"effectively":[6],"optimized":[7,49,72],"remains":[8,146],"challenging,":[9],"as":[10],"training":[11,37,63,154,182],"occurs":[12],"in":[13,44,65,136],"highly":[14],"nonconvex":[15],"landscapes":[16],"and":[17,47,86,150],"standard":[18],"metrics":[19],"provide":[20],"limited":[21],"visibility":[22],"into":[23],"layer-wise":[24,58,115],"learning":[25],"quality.":[26],"This":[27],"challenge":[28],"is":[29,38,70],"particularly":[30,157],"acute":[31],"for":[32,61],"transformer-based":[33],"language":[34],"models,":[35],"where":[36,153],"expensive,":[39],"models":[40,111],"often":[42],"reused":[43],"frozen":[45],"form,":[46],"poorly":[48],"layers":[50,88],"can":[51,118],"silently":[52],"degrade":[53],"performance.":[54],"We":[55,140],"propose":[56],"a":[57],"peeling":[59],"framework":[60],"monitoring":[62],"dynamics,":[64],"which":[66],"each":[67],"transformer":[68,110],"layer":[69],"locally":[71],"against":[73],"intermediate":[74,91],"representations":[75],"of":[76,104,129],"the":[77,123,163],"trained":[78,124],"model.":[79],"By":[80],"constructing":[81],"lightweight,":[82],"layer-specific":[83],"reference":[84,116],"solutions":[85],"projecting":[87],"onto":[89],"multiple":[90],"outputs":[92],"via":[93],"different":[94],"permutations,":[95],"we":[96],"obtain":[97],"achievable":[98],"baselines":[99],"that":[100,113,133,143,176],"enable":[101],"fine-grained":[102],"diagnosis":[103],"under-optimized":[105],"layers.":[106],"Experiments":[107],"on":[108,181],"decoder-only":[109],"show":[112],"these":[114],"bounds":[117,165],"match":[119],"or":[120],"even":[121],"surpass":[122],"model":[125],"at":[126],"various":[127],"stages":[128],"training,":[130],"exposing":[131],"inefficiencies":[132],"remain":[134],"hidden":[135],"aggregate":[137],"loss":[138,183],"curves.":[139],"further":[141],"demonstrate":[142],"this":[144],"analysis":[145],"effective":[147,171],"under":[148],"binarization":[149],"quantized":[151],"settings,":[152],"dynamics":[155],"fragile.":[158],"Across":[159],"all":[160],"numerical":[161],"results,":[162],"proposed":[164],"consistently":[166],"separate":[167],"apparent":[168],"convergence":[169],"from":[170],"optimality,":[172],"highlighting":[173],"optimization":[174],"opportunities":[175],"invisible":[178],"when":[179],"relying":[180],"alone.":[184]},"counts_by_year":[],"updated_date":"2026-06-11T09:08:48.828518","created_date":"2026-05-06T00:00:00"}