{"id":"https://openalex.org/W7164003924","doi":"https://doi.org/10.48550/arxiv.2606.08985","title":"Beyond Neural Collapse: Task-Intrinsic Geometry Governs Neural Representations in Modular Arithmetic","display_name":"Beyond Neural Collapse: Task-Intrinsic Geometry Governs Neural Representations in Modular Arithmetic","publication_year":2026,"publication_date":"2026-06-08","ids":{"openalex":"https://openalex.org/W7164003924","doi":"https://doi.org/10.48550/arxiv.2606.08985"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2606.08985","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.08985","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":null,"license_id":null,"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.2606.08985","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5138219629","display_name":"Hu Tan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Tan, Hu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5138248938","display_name":"Kuo Gai","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Gai, Kuo","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5076619069","display_name":"Shihua Zhang","orcid":"https://orcid.org/0000-0003-0192-7118"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Shihua","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/T12808","display_name":"Ferroelectric and Negative Capacitance Devices","score":0.21170000731945038,"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"}},"topics":[{"id":"https://openalex.org/T12808","display_name":"Ferroelectric and Negative Capacitance Devices","score":0.21170000731945038,"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"}},{"id":"https://openalex.org/T11094","display_name":"Face Recognition and Perception","score":0.15039999783039093,"subfield":{"id":"https://openalex.org/subfields/2805","display_name":"Cognitive Neuroscience"},"field":{"id":"https://openalex.org/fields/28","display_name":"Neuroscience"},"domain":{"id":"https://openalex.org/domains/1","display_name":"Life Sciences"}},{"id":"https://openalex.org/T10581","display_name":"Neural dynamics and brain function","score":0.0689999982714653,"subfield":{"id":"https://openalex.org/subfields/2805","display_name":"Cognitive Neuroscience"},"field":{"id":"https://openalex.org/fields/28","display_name":"Neuroscience"},"domain":{"id":"https://openalex.org/domains/1","display_name":"Life Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/classifier","display_name":"Classifier (UML)","score":0.6680999994277954},{"id":"https://openalex.org/keywords/embedding","display_name":"Embedding","score":0.5978000164031982},{"id":"https://openalex.org/keywords/simplex","display_name":"Simplex","score":0.5561000108718872},{"id":"https://openalex.org/keywords/subspace-topology","display_name":"Subspace topology","score":0.5550000071525574},{"id":"https://openalex.org/keywords/modular-design","display_name":"Modular design","score":0.5454999804496765},{"id":"https://openalex.org/keywords/artificial-neural-network","display_name":"Artificial neural network","score":0.4569000005722046},{"id":"https://openalex.org/keywords/discriminant","display_name":"Discriminant","score":0.3871999979019165},{"id":"https://openalex.org/keywords/pattern-recognition","display_name":"Pattern recognition (psychology)","score":0.33559998869895935}],"concepts":[{"id":"https://openalex.org/C95623464","wikidata":"https://www.wikidata.org/wiki/Q1096149","display_name":"Classifier (UML)","level":2,"score":0.6680999994277954},{"id":"https://openalex.org/C41608201","wikidata":"https://www.wikidata.org/wiki/Q980509","display_name":"Embedding","level":2,"score":0.5978000164031982},{"id":"https://openalex.org/C62438384","wikidata":"https://www.wikidata.org/wiki/Q331350","display_name":"Simplex","level":2,"score":0.5561000108718872},{"id":"https://openalex.org/C32834561","wikidata":"https://www.wikidata.org/wiki/Q660730","display_name":"Subspace topology","level":2,"score":0.5550000071525574},{"id":"https://openalex.org/C101468663","wikidata":"https://www.wikidata.org/wiki/Q1620158","display_name":"Modular design","level":2,"score":0.5454999804496765},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.545199990272522},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.4731000065803528},{"id":"https://openalex.org/C50644808","wikidata":"https://www.wikidata.org/wiki/Q192776","display_name":"Artificial neural network","level":2,"score":0.4569000005722046},{"id":"https://openalex.org/C78397625","wikidata":"https://www.wikidata.org/wiki/Q192487","display_name":"Discriminant","level":2,"score":0.3871999979019165},{"id":"https://openalex.org/C2524010","wikidata":"https://www.wikidata.org/wiki/Q8087","display_name":"Geometry","level":1,"score":0.3529999852180481},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.3522000014781952},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.34950000047683716},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.33559998869895935},{"id":"https://openalex.org/C178009071","wikidata":"https://www.wikidata.org/wiki/Q93344","display_name":"Trigonometric functions","level":2,"score":0.3239000141620636},{"id":"https://openalex.org/C96469262","wikidata":"https://www.wikidata.org/wiki/Q1324364","display_name":"Homogeneous space","level":2,"score":0.31949999928474426},{"id":"https://openalex.org/C184720557","wikidata":"https://www.wikidata.org/wiki/Q7825049","display_name":"Topology (electrical circuits)","level":2,"score":0.31540000438690186},{"id":"https://openalex.org/C83665646","wikidata":"https://www.wikidata.org/wiki/Q42139305","display_name":"Feature vector","level":2,"score":0.31279999017715454},{"id":"https://openalex.org/C111030470","wikidata":"https://www.wikidata.org/wiki/Q1430460","display_name":"Curse of dimensionality","level":2,"score":0.302700012922287},{"id":"https://openalex.org/C29001434","wikidata":"https://www.wikidata.org/wiki/Q8084","display_name":"Trigonometry","level":2,"score":0.2904999852180481},{"id":"https://openalex.org/C176809094","wikidata":"https://www.wikidata.org/wiki/Q15401496","display_name":"Traverse","level":2,"score":0.2782999873161316},{"id":"https://openalex.org/C205711294","wikidata":"https://www.wikidata.org/wiki/Q176953","display_name":"Rendering (computer graphics)","level":2,"score":0.2768000066280365},{"id":"https://openalex.org/C74261601","wikidata":"https://www.wikidata.org/wiki/Q40112","display_name":"Ellipse","level":2,"score":0.2603999972343445}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2606.08985","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.08985","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2606.08985","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.08985","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":null,"license_id":null,"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":{"While":[0],"neural":[1],"collapse":[2],"(NC)":[3],"predicts":[4],"that":[5,203],"a":[6,14,25,31,59,74,144,155,171,180,217],"$K$-class-balanced":[7],"classifier":[8,38,65,86,191],"should":[9],"organize":[10],"terminal":[11],"representations":[12],"as":[13],"$(K-1)$-dimensional":[15],"simplex":[16,156],"equiangular":[17,76],"tight":[18],"frame":[19],"(ETF),":[20],"modular":[21,206],"addition":[22],"consistently":[23],"enters":[24],"different":[26],"regime:":[27],"networks":[28],"compress":[29],"to":[30,95,129],"two-dimensional":[32],"cyclic":[33,167],"geometry":[34],"in":[35,53,163],"which":[36],"both":[37,189],"weights":[39,66,192],"and":[40,83,139,195,223],"token":[41],"embeddings":[42,80,197],"lie":[43],"on":[44,119,143,205],"circles.":[45],"We":[46],"refine":[47],"the":[48,96,110,166],"explanation":[49],"of":[50,137],"this":[51,85,107,125,150],"phenomenon":[52],"three":[54],"directions.":[55],"First,":[56],"we":[57,147],"formalize":[58],"layerwise":[60],"non-uniform":[61],"training":[62],"mechanism:":[63],"downstream":[64],"are":[67,134],"driven":[68],"by":[69,211,216],"dense":[70],"cross-entropy":[71],"gradients":[72,91],"into":[73],"rank-2":[75,168],"configuration":[77],"before":[78],"upstream":[79],"fully":[81],"reorganize,":[82],"once":[84],"plane":[87,98],"forms,":[88],"backpropagated":[89],"feature":[90],"constrain":[92],"embedding":[93,127],"motion":[94],"same":[97],"while":[99],"weight":[100],"decay":[101],"suppresses":[102],"orthogonal":[103],"components.":[104],"Second,":[105],"after":[106],"subspace":[108],"locking,":[109],"induced":[111],"in-plane":[112],"dynamics":[113],"admit":[114],"an":[115,160],"entropy-regularized":[116],"transport":[117],"interpretation":[118],"$S^1$;":[120],"combined":[121],"with":[122,200],"modular-addition":[123],"labels,":[124],"reduces":[126],"formation":[128],"phase":[130],"alignment,":[131],"whose":[132],"minimizers":[133],"single-frequency":[135],"characters":[136],"$\\mathbb{Z}/P\\mathbb{Z}$":[138],"hence":[140],"equal-angle":[141],"points":[142],"circle.":[145],"Third,":[146],"quantify":[148],"why":[149,190,196],"solution":[151,169],"prevails":[152],"over":[153],"NC:":[154],"ETF":[157],"gains":[158],"only":[159],"$O(1)$":[161],"advantage":[162,173],"cross-entropy,":[164],"whereas":[165],"enjoys":[170],"$\u0398(K)$":[172],"under":[174],"Schatten":[175],"or":[176],"weight-decay":[177],"surrogates,":[178],"yielding":[179],"critical":[181],"threshold":[182],"$\u03bb_{\\mathrm{crit}}":[183],"=":[184],"\u0398(1/K)$.":[185],"Our":[186],"results":[187],"explain":[188],"move":[193],"first":[194],"subsequently":[198],"align":[199],"them,":[201],"showing":[202],"grokking":[204],"arithmetic":[207],"is":[208],"governed":[209],"not":[210],"maximal":[212],"separation":[213],"alone":[214],"but":[215],"task-structured":[218],"trade-off":[219],"between":[220],"separation,":[221],"symmetry,":[222],"complexity.":[224]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-06-10T00:00:00"}