{"id":"https://openalex.org/W4387947192","doi":"https://doi.org/10.48550/arxiv.2310.15440","title":"Learning Dynamics in Linear VAE: Posterior Collapse Threshold, Superfluous Latent Space Pitfalls, and Speedup with KL Annealing","display_name":"Learning Dynamics in Linear VAE: Posterior Collapse Threshold, Superfluous Latent Space Pitfalls, and Speedup with KL Annealing","publication_year":2023,"publication_date":"2023-10-24","ids":{"openalex":"https://openalex.org/W4387947192","doi":"https://doi.org/10.48550/arxiv.2310.15440"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2310.15440","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2310.15440","pdf_url":"https://arxiv.org/pdf/2310.15440","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","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2310.15440","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5075950477","display_name":"Yuma Ichikawa","orcid":"https://orcid.org/0009-0004-4216-7017","observed_orcids":["https://orcid.org/0009-0004-4216-7017"]},"institutions":[{"id":"https://openalex.org/I2252096349","display_name":"Fujitsu (Japan)","ror":"https://ror.org/038e2g226","country_code":"JP","type":"company","lineage":["https://openalex.org/I2252096349"]},{"id":"https://openalex.org/I74801974","display_name":"The University of Tokyo","ror":"https://ror.org/057zh3y96","country_code":"JP","type":"education","lineage":["https://openalex.org/I74801974"]}],"countries":["JP"],"is_corresponding":false,"raw_author_name":"Ichikawa, Yuma","raw_affiliation_strings":["University of Tokyo","University of Tokyo , Fujitsu Limited "],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Tokyo","institution_ids":["https://openalex.org/I74801974"]},{"raw_affiliation_string":"University of Tokyo , Fujitsu Limited ","institution_ids":["https://openalex.org/I2252096349","https://openalex.org/I74801974"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5085397996","display_name":"Koji Hukushima","orcid":"https://orcid.org/0000-0003-1153-1758","observed_orcids":["https://orcid.org/0000-0003-1153-1758"]},"institutions":[{"id":"https://openalex.org/I2252096349","display_name":"Fujitsu (Japan)","ror":"https://ror.org/038e2g226","country_code":"JP","type":"company","lineage":["https://openalex.org/I2252096349"]},{"id":"https://openalex.org/I74801974","display_name":"The University of Tokyo","ror":"https://ror.org/057zh3y96","country_code":"JP","type":"education","lineage":["https://openalex.org/I74801974"]}],"countries":["JP"],"is_corresponding":false,"raw_author_name":"Hukushima, Koji","raw_affiliation_strings":["University of Tokyo","University of Tokyo , Fujitsu Limited "],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Tokyo","institution_ids":["https://openalex.org/I74801974"]},{"raw_affiliation_string":"University of Tokyo , Fujitsu Limited ","institution_ids":["https://openalex.org/I2252096349","https://openalex.org/I74801974"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":2,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":4,"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/T12814","display_name":"Gaussian Processes and Bayesian Inference","score":0.5913000106811523,"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/T12814","display_name":"Gaussian Processes and Bayesian Inference","score":0.5913000106811523,"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":0.11219999939203262,"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/T11612","display_name":"Stochastic Gradient Optimization Techniques","score":0.07270000129938126,"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/variational-autoencoder","display_name":"variational autoencoder","score":0.9020000100135803},{"id":"https://openalex.org/keywords/posterior-collapse","display_name":"posterior collapse","score":0.996999979019165},{"id":"https://openalex.org/keywords/learning-dynamics","display_name":"learning dynamics","score":0.8889999985694885},{"id":"https://openalex.org/keywords/disentangled-representations","display_name":"disentangled representations","score":0.7440000176429749},{"id":"https://openalex.org/keywords/generalization-error","display_name":"generalization error","score":0.753000020980835},{"id":"https://openalex.org/keywords/fixed-point-analysis","display_name":"fixed-point analysis","score":0.8759999871253967},{"id":"https://openalex.org/keywords/deterministic-dynamics","display_name":"deterministic dynamics","score":0.3889999985694885},{"id":"https://openalex.org/keywords/convergence-acceleration","display_name":"convergence acceleration","score":0.4000000059604645}],"concepts":[{"id":"https://openalex.org/C22019652","wikidata":"https://www.wikidata.org/wiki/Q331309","display_name":"Overfitting","level":3,"score":0.8508331775665283},{"id":"https://openalex.org/C8642999","wikidata":"https://www.wikidata.org/wiki/Q4171168","display_name":"Hyperparameter","level":2,"score":0.7231899499893188},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6479546427726746},{"id":"https://openalex.org/C68339613","wikidata":"https://www.wikidata.org/wiki/Q1549489","display_name":"Speedup","level":2,"score":0.5179204344749451},{"id":"https://openalex.org/C177148314","wikidata":"https://www.wikidata.org/wiki/Q170084","display_name":"Generalization","level":2,"score":0.46451014280319214},{"id":"https://openalex.org/C167966045","wikidata":"https://www.wikidata.org/wiki/Q5532625","display_name":"Generative model","level":3,"score":0.429721474647522},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.41839081048965454},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.38790860772132874},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.3519914150238037},{"id":"https://openalex.org/C39890363","wikidata":"https://www.wikidata.org/wiki/Q36108","display_name":"Generative grammar","level":2,"score":0.3479483127593994},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.337636798620224},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.2426687479019165},{"id":"https://openalex.org/C50644808","wikidata":"https://www.wikidata.org/wiki/Q192776","display_name":"Artificial neural network","level":2,"score":0.19372230768203735},{"id":"https://openalex.org/C111919701","wikidata":"https://www.wikidata.org/wiki/Q9135","display_name":"Operating system","level":1,"score":0.0},{"id":"https://openalex.org/C134306372","wikidata":"https://www.wikidata.org/wiki/Q7754","display_name":"Mathematical analysis","level":1,"score":0.0}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2310.15440","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2310.15440","pdf_url":"https://arxiv.org/pdf/2310.15440","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.2310.15440","doi":"https://doi.org/10.48550/arxiv.2310.15440","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2310.15440","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2310.15440","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2310.15440","pdf_url":"https://arxiv.org/pdf/2310.15440","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":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4387947192.pdf","grobid_xml":"https://content.openalex.org/works/W4387947192.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4380136386","https://openalex.org/W4386794890","https://openalex.org/W7167680863","https://openalex.org/W4414817983","https://openalex.org/W2910135751","https://openalex.org/W2995875017","https://openalex.org/W2983437672","https://openalex.org/W4384526194","https://openalex.org/W3131804914","https://openalex.org/W6948181718"],"abstract_inverted_index":{"Variational":[0],"autoencoders":[1],"(VAEs)":[2],"face":[3],"a":[4,17,39,53,61,73,85,121],"notorious":[5],"problem":[6],"wherein":[7],"the":[8,15,25,57,69,77,90,94,98,113,130,134,139,146],"variational":[9],"posterior":[10,21,124],"often":[11],"aligns":[12],"closely":[13],"with":[14],"prior,":[16],"phenomenon":[18],"known":[19],"as":[20],"collapse,":[22],"which":[23],"hinders":[24],"quality":[26],"of":[27,56,79,89,112,129,145],"representation":[28],"learning.":[29],"To":[30],"mitigate":[31],"this":[32,43,149],"problem,":[33],"an":[34],"adjustable":[35],"hyperparameter":[36],"$\u03b2$":[37,119],"and":[38,104,154],"strategy":[40],"for":[41,138],"annealing":[42,165],"parameter,":[44],"called":[45],"KL":[46,164],"annealing,":[47],"are":[48],"proposed.":[49],"This":[50],"study":[51],"presents":[52],"theoretical":[54],"analysis":[55,88,95,111,158],"learning":[58,131,155],"dynamics":[59,70],"in":[60],"minimal":[62],"VAE.":[63],"It":[64],"is":[65],"rigorously":[66],"proved":[67],"that":[68,97,117,161],"converge":[71],"to":[72,143],"deterministic":[74,114],"process":[75,115],"within":[76],"limit":[78],"large":[80],"input":[81],"dimensions,":[82],"thereby":[83],"enabling":[84],"detailed":[86],"dynamical":[87],"generalization":[91,153],"error.":[92],"Furthermore,":[93],"shows":[96],"VAE":[99],"initially":[100],"learns":[101],"entangled":[102],"representations":[103],"gradually":[105],"acquires":[106],"disentangled":[107],"representations.":[108],"A":[109],"fixed-point":[110],"reveals":[116],"when":[118],"exceeds":[120],"certain":[122],"threshold,":[123],"collapse":[125],"becomes":[126],"inevitable":[127],"regardless":[128],"period.":[132],"Additionally,":[133],"superfluous":[135],"latent":[136],"variables":[137],"data-generative":[140],"factors":[141],"lead":[142],"overfitting":[144],"background":[147],"noise;":[148],"adversely":[150],"affects":[151],"both":[152],"convergence.":[156,168],"The":[157],"further":[159],"unveiled":[160],"appropriately":[162],"tuned":[163],"can":[166],"accelerate":[167]},"counts_by_year":[{"year":2026,"cited_by_count":1},{"year":2025,"cited_by_count":1},{"year":2024,"cited_by_count":2}],"updated_date":"2026-10-16T09:59:22.235173","created_date":"2023-10-26T00:00:00"}