Disentanglement by nonlinear ICA with General Incompressible-flow Networks (GIN)

A central question of representation learning asks under which conditions it is possible to reconstruct the true latent variables of an arbitrarily complex generative process. Recent breakthrough work by Khemakhem et al. (2019) on nonlinear ICA has answered this question for a broad class of conditi...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Sorrenson, Peter (VerfasserIn) , Rother, Carsten (VerfasserIn) , Köthe, Ullrich (VerfasserIn)
Dokumenttyp: Article (Journal) Kapitel/Artikel
Sprache:Englisch
Veröffentlicht: 14 Jan 2020
In: Arxiv
Year: 2020, Pages: 1-23
DOI:10.48550/arXiv.2001.04872
Online-Zugang:Verlag, lizenzpflichtig, Volltext: https://doi.org/10.48550/arXiv.2001.04872
Verlag, lizenzpflichtig, Volltext: http://arxiv.org/abs/2001.04872
Volltext
Verfasserangaben:Peter Sorrenson, Carsten Rother, Ullrich Köthe

MARC

LEADER 00000caa a2200000 c 4500
001 1810850444
003 DE-627
005 20220820225255.0
007 cr uuu---uuuuu
008 220719s2020 xx |||||o 00| ||eng c
024 7 |a 10.48550/arXiv.2001.04872  |2 doi 
035 |a (DE-627)1810850444 
035 |a (DE-599)KXP1810850444 
035 |a (OCoLC)1341463921 
040 |a DE-627  |b ger  |c DE-627  |e rda 
041 |a eng 
084 |a 28  |2 sdnb 
100 1 |a Sorrenson, Peter  |d 1994-  |e VerfasserIn  |0 (DE-588)1224419480  |0 (DE-627)1743772106  |4 aut 
245 1 0 |a Disentanglement by nonlinear ICA with General Incompressible-flow Networks (GIN)  |c Peter Sorrenson, Carsten Rother, Ullrich Köthe 
264 1 |c 14 Jan 2020 
300 |a 23 
336 |a Text  |b txt  |2 rdacontent 
337 |a Computermedien  |b c  |2 rdamedia 
338 |a Online-Ressource  |b cr  |2 rdacarrier 
500 |a Gesehen am 19.07.2022 
520 |a A central question of representation learning asks under which conditions it is possible to reconstruct the true latent variables of an arbitrarily complex generative process. Recent breakthrough work by Khemakhem et al. (2019) on nonlinear ICA has answered this question for a broad class of conditional generative processes. We extend this important result in a direction relevant for application to real-world data. First, we generalize the theory to the case of unknown intrinsic problem dimension and prove that in some special (but not very restrictive) cases, informative latent variables will be automatically separated from noise by an estimating model. Furthermore, the recovered informative latent variables will be in one-to-one correspondence with the true latent variables of the generating process, up to a trivial component-wise transformation. Second, we introduce a modification of the RealNVP invertible neural network architecture (Dinh et al. (2016)) which is particularly suitable for this type of problem: the General Incompressible-flow Network (GIN). Experiments on artificial data and EMNIST demonstrate that theoretical predictions are indeed verified in practice. In particular, we provide a detailed set of exactly 22 informative latent variables extracted from EMNIST. 
650 4 |a Computer Science - Machine Learning 
650 4 |a Statistics - Machine Learning 
700 1 |a Rother, Carsten  |e VerfasserIn  |0 (DE-588)1181464692  |0 (DE-627)1662676883  |4 aut 
700 1 |a Köthe, Ullrich  |e VerfasserIn  |0 (DE-588)123963435  |0 (DE-627)594480884  |0 (DE-576)304484520  |4 aut 
773 0 8 |i Enthalten in  |t Arxiv  |d Ithaca, NY : Cornell University, 1991  |g (2020), Artikel-ID 2001.04872, Seite 1-23  |h Online-Ressource  |w (DE-627)509006531  |w (DE-600)2225896-6  |w (DE-576)28130436X  |7 nnas  |a Disentanglement by nonlinear ICA with General Incompressible-flow Networks (GIN) 
773 1 8 |g year:2020  |g elocationid:2001.04872  |g pages:1-23  |g extent:23  |a Disentanglement by nonlinear ICA with General Incompressible-flow Networks (GIN) 
856 4 0 |u https://doi.org/10.48550/arXiv.2001.04872  |x Verlag  |x Resolving-System  |z lizenzpflichtig  |3 Volltext 
856 4 0 |u http://arxiv.org/abs/2001.04872  |x Verlag  |z lizenzpflichtig  |3 Volltext 
951 |a AR 
992 |a 20220719 
993 |a Article 
994 |a 2020 
998 |g 123963435  |a Köthe, Ullrich  |m 123963435:Köthe, Ullrich  |d 700000  |d 708070  |d 700000  |d 728500  |e 700000PK123963435  |e 708070PK123963435  |e 700000PK123963435  |e 728500PK123963435  |k 0/700000/  |k 1/700000/708070/  |k 0/700000/  |k 1/700000/728500/  |p 3  |y j 
998 |g 1181464692  |a Rother, Carsten  |m 1181464692:Rother, Carsten  |d 700000  |d 708070  |d 700000  |d 728500  |e 700000PR1181464692  |e 708070PR1181464692  |e 700000PR1181464692  |e 728500PR1181464692  |k 0/700000/  |k 1/700000/708070/  |k 0/700000/  |k 1/700000/728500/  |p 2 
998 |g 1224419480  |a Sorrenson, Peter  |m 1224419480:Sorrenson, Peter  |d 700000  |d 708070  |e 700000PS1224419480  |e 708070PS1224419480  |k 0/700000/  |k 1/700000/708070/  |p 1  |x j 
999 |a KXP-PPN1810850444  |e 4170218353 
BIB |a Y 
JSO |a {"title":[{"title":"Disentanglement by nonlinear ICA with General Incompressible-flow Networks (GIN)","title_sort":"Disentanglement by nonlinear ICA with General Incompressible-flow Networks (GIN)"}],"person":[{"display":"Sorrenson, Peter","roleDisplay":"VerfasserIn","role":"aut","family":"Sorrenson","given":"Peter"},{"role":"aut","display":"Rother, Carsten","roleDisplay":"VerfasserIn","given":"Carsten","family":"Rother"},{"roleDisplay":"VerfasserIn","display":"Köthe, Ullrich","role":"aut","family":"Köthe","given":"Ullrich"}],"language":["eng"],"recId":"1810850444","note":["Gesehen am 19.07.2022"],"type":{"media":"Online-Ressource","bibl":"chapter"},"id":{"eki":["1810850444"],"doi":["10.48550/arXiv.2001.04872"]},"origin":[{"dateIssuedKey":"2020","dateIssuedDisp":"14 Jan 2020"}],"name":{"displayForm":["Peter Sorrenson, Carsten Rother, Ullrich Köthe"]},"relHost":[{"title":[{"title":"Arxiv","title_sort":"Arxiv"}],"language":["eng"],"recId":"509006531","disp":"Disentanglement by nonlinear ICA with General Incompressible-flow Networks (GIN)Arxiv","type":{"bibl":"edited-book","media":"Online-Ressource"},"note":["Gesehen am 28.05.2024"],"titleAlt":[{"title":"Arxiv.org"},{"title":"Arxiv.org e-print archive"},{"title":"Arxiv e-print archive"},{"title":"De.arxiv.org"}],"part":{"extent":"23","text":"(2020), Artikel-ID 2001.04872, Seite 1-23","pages":"1-23","year":"2020"},"pubHistory":["1991 -"],"id":{"zdb":["2225896-6"],"eki":["509006531"]},"origin":[{"publisher":"Cornell University ; Arxiv.org","dateIssuedKey":"1991","dateIssuedDisp":"1991-","publisherPlace":"Ithaca, NY ; [Erscheinungsort nicht ermittelbar]"}],"physDesc":[{"extent":"Online-Ressource"}]}],"physDesc":[{"extent":"23 S."}]} 
SRT |a SORRENSONPDISENTANGL1420