{"id":"https://openalex.org/W7152083639","doi":"https://doi.org/10.48550/arxiv.2604.05414","title":"Training Without Orthogonalization, Inference With SVD: A Gradient Analysis of Rotation Representations","display_name":"Training Without Orthogonalization, Inference With SVD: A Gradient Analysis of Rotation Representations","publication_year":2026,"publication_date":"2026-04-07","ids":{"openalex":"https://openalex.org/W7152083639","doi":"https://doi.org/10.48550/arxiv.2604.05414"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2604.05414","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.05414","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.2604.05414","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5133215132","display_name":"Chris Choy","orcid":null},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Choy, Chris","raw_affiliation_strings":[],"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":1,"corresponding_author_ids":["https://openalex.org/A5133215132"],"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/T11612","display_name":"Stochastic Gradient Optimization Techniques","score":0.5595999956130981,"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":0.5595999956130981,"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/T12814","display_name":"Gaussian Processes and Bayesian Inference","score":0.039000000804662704,"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/T10500","display_name":"Sparse and Compressive Sensing Techniques","score":0.026799999177455902,"subfield":{"id":"https://openalex.org/subfields/2206","display_name":"Computational Mechanics"},"field":{"id":"https://openalex.org/fields/22","display_name":"Engineering"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/orthogonalization","display_name":"Orthogonalization","score":0.859000027179718},{"id":"https://openalex.org/keywords/singular-value-decomposition","display_name":"Singular value decomposition","score":0.8483999967575073},{"id":"https://openalex.org/keywords/projection","display_name":"Projection (relational algebra)","score":0.5751000046730042},{"id":"https://openalex.org/keywords/jacobian-matrix-and-determinant","display_name":"Jacobian matrix and determinant","score":0.5609999895095825},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.5318999886512756},{"id":"https://openalex.org/keywords/singular-value","display_name":"Singular value","score":0.5242000222206116},{"id":"https://openalex.org/keywords/dimension","display_name":"Dimension (graph theory)","score":0.517799973487854},{"id":"https://openalex.org/keywords/gradient-descent","display_name":"Gradient descent","score":0.4410000145435333},{"id":"https://openalex.org/keywords/matrix","display_name":"Matrix (chemical analysis)","score":0.4327999949455261}],"concepts":[{"id":"https://openalex.org/C47559304","wikidata":"https://www.wikidata.org/wiki/Q1702189","display_name":"Orthogonalization","level":2,"score":0.859000027179718},{"id":"https://openalex.org/C22789450","wikidata":"https://www.wikidata.org/wiki/Q420904","display_name":"Singular value decomposition","level":2,"score":0.8483999967575073},{"id":"https://openalex.org/C57493831","wikidata":"https://www.wikidata.org/wiki/Q3134666","display_name":"Projection (relational algebra)","level":2,"score":0.5751000046730042},{"id":"https://openalex.org/C200331156","wikidata":"https://www.wikidata.org/wiki/Q506041","display_name":"Jacobian matrix and determinant","level":2,"score":0.5609999895095825},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.5480999946594238},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.5318999886512756},{"id":"https://openalex.org/C109282560","wikidata":"https://www.wikidata.org/wiki/Q4166054","display_name":"Singular value","level":3,"score":0.5242000222206116},{"id":"https://openalex.org/C33676613","wikidata":"https://www.wikidata.org/wiki/Q13415176","display_name":"Dimension (graph theory)","level":2,"score":0.517799973487854},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.48399999737739563},{"id":"https://openalex.org/C153258448","wikidata":"https://www.wikidata.org/wiki/Q1199743","display_name":"Gradient descent","level":3,"score":0.4410000145435333},{"id":"https://openalex.org/C106487976","wikidata":"https://www.wikidata.org/wiki/Q685816","display_name":"Matrix (chemical analysis)","level":2,"score":0.4327999949455261},{"id":"https://openalex.org/C164226766","wikidata":"https://www.wikidata.org/wiki/Q7293202","display_name":"Rank (graph theory)","level":2,"score":0.4221999943256378},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4131999909877777},{"id":"https://openalex.org/C74050887","wikidata":"https://www.wikidata.org/wiki/Q848368","display_name":"Rotation (mathematics)","level":2,"score":0.37049999833106995},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.3695000112056732},{"id":"https://openalex.org/C158693339","wikidata":"https://www.wikidata.org/wiki/Q190524","display_name":"Eigenvalues and eigenvectors","level":2,"score":0.36469998955726624},{"id":"https://openalex.org/C28826006","wikidata":"https://www.wikidata.org/wiki/Q33521","display_name":"Applied mathematics","level":1,"score":0.33869999647140503},{"id":"https://openalex.org/C175694140","wikidata":"https://www.wikidata.org/wiki/Q980329","display_name":"Orthographic projection","level":2,"score":0.33709999918937683},{"id":"https://openalex.org/C115680565","wikidata":"https://www.wikidata.org/wiki/Q5977448","display_name":"Gradient method","level":2,"score":0.328000009059906},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.32690000534057617},{"id":"https://openalex.org/C137800194","wikidata":"https://www.wikidata.org/wiki/Q11713455","display_name":"Interpolation (computer graphics)","level":3,"score":0.3140000104904175},{"id":"https://openalex.org/C39482219","wikidata":"https://www.wikidata.org/wiki/Q192826","display_name":"Kronecker delta","level":2,"score":0.3098999857902527},{"id":"https://openalex.org/C42355184","wikidata":"https://www.wikidata.org/wiki/Q1361088","display_name":"Matrix decomposition","level":3,"score":0.3059999942779541},{"id":"https://openalex.org/C77246614","wikidata":"https://www.wikidata.org/wiki/Q1409400","display_name":"Gramian matrix","level":3,"score":0.30410000681877136},{"id":"https://openalex.org/C90199385","wikidata":"https://www.wikidata.org/wiki/Q6692777","display_name":"Low-rank approximation","level":3,"score":0.2903999984264374},{"id":"https://openalex.org/C70518039","wikidata":"https://www.wikidata.org/wiki/Q16000077","display_name":"Dimensionality reduction","level":2,"score":0.2736999988555908},{"id":"https://openalex.org/C169756996","wikidata":"https://www.wikidata.org/wiki/Q194919","display_name":"Eigendecomposition of a matrix","level":3,"score":0.2628999948501587},{"id":"https://openalex.org/C50644808","wikidata":"https://www.wikidata.org/wiki/Q192776","display_name":"Artificial neural network","level":2,"score":0.2574999928474426}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2604.05414","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.05414","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.2604.05414","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.05414","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":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Recent":[0],"work":[1],"has":[2,85,164],"shown":[3],"that":[4,113,136,159],"removing":[5,146],"orthogonalization":[6,37,61],"during":[7],"training":[8,128,150,188],"and":[9,41,68,100,193],"applying":[10,194],"it":[11,43,84],"only":[12,197],"at":[13,49,198],"inference":[14],"improves":[15],"rotation":[16],"estimation":[17],"in":[18,127],"deep":[19],"learning,":[20],"with":[21,27,93,189],"empirical":[22],"evidence":[23],"favoring":[24],"9D":[25,176,191],"representations":[26],"SVD":[28,36,60,80,139,147,195],"projection.":[29,70],"However,":[30],"the":[31,75,79,89,118,149,160,184],"theoretical":[32,185],"understanding":[33],"of":[34,59,78,91],"why":[35,42,175],"specifically":[38],"harms":[39],"training,":[40],"should":[44],"be":[45],"preferred":[46],"over":[47],"Gram-Schmidt":[48,162],"inference,":[50],"remains":[51],"incomplete.":[52],"We":[53,133,156],"provide":[54,183],"a":[55],"detailed":[56],"gradient":[57,111,142,172],"analysis":[58],"specialized":[62],"to":[63],"$3":[64],"\\times":[65],"3$":[66],"matrices":[67],"$SO(3)$":[69,124],"Our":[71],"central":[72],"result":[73],"derives":[74],"exact":[76],"spectrum":[77],"backward":[81],"pass":[82],"Jacobian:":[83],"rank":[86],"$3$":[87],"(matching":[88],"dimension":[90],"$SO(3)$)":[92],"nonzero":[94],"singular":[95],"values":[96],"$2/(s_i":[97],"+":[98,105,107],"s_j)$":[99],"condition":[101],"number":[102],"$\u03ba=":[103],"(s_1":[104],"s_2)/(s_2":[106],"s_3)$,":[108],"creating":[109],"quantifiable":[110],"distortion":[112],"is":[114,121,178],"most":[115],"severe":[116],"when":[117,129],"predicted":[119],"matrix":[120],"far":[122],"from":[123,148],"(e.g.,":[125],"early":[126],"$s_3":[130],"\\approx":[131],"0$).":[132],"further":[134],"show":[135],"even":[137],"stabilized":[138],"gradients":[140],"introduce":[141],"direction":[143],"error,":[144],"whereas":[145],"loop":[151],"avoids":[152],"this":[153],"tradeoff":[154],"entirely.":[155],"also":[157],"prove":[158],"6D":[161],"Jacobian":[163],"an":[165],"asymmetric":[166],"spectrum:":[167],"its":[168],"parameters":[169],"receive":[170],"unequal":[171],"signal,":[173],"explaining":[174],"parameterization":[177],"preferable.":[179],"Together,":[180],"these":[181],"results":[182],"foundation":[186],"for":[187],"direct":[190],"regression":[192],"projection":[196],"inference.":[199]},"counts_by_year":[],"updated_date":"2026-04-09T06:13:59.934233","created_date":"2026-04-09T00:00:00"}