Generative assignment flows for representing and learning joint distributions of discrete data

We introduce a novel generative model for the representation of joint probability distributions of a possibly large number of discrete random variables. The approach uses measure transport by randomized assignment flows on the statistical submanifold of factorizing distributions, which enables to re...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Boll, Bastian (VerfasserIn) , Gonzalez Alvarado, Daniel (VerfasserIn) , Petra, Stefania (VerfasserIn) , Schnörr, Christoph (VerfasserIn)
Dokumenttyp: Article (Journal)
Sprache:Englisch
Veröffentlicht: 28 May 2025
In: Journal of mathematical imaging and vision
Year: 2025, Jahrgang: 67, Heft: 3, Pages: 1-24
ISSN:1573-7683
DOI:10.1007/s10851-025-01239-9
Online-Zugang:Verlag, kostenfrei, Volltext: https://doi.org/10.1007/s10851-025-01239-9
Verlag, kostenfrei, Volltext: https://link.springer.com/article/10.1007/s10851-025-01239-9
Volltext
Verfasserangaben:Bastian Boll, Daniel Gonzalez-Alvarado, Stefania Petra, Christoph Schnörr

MARC

LEADER 00000naa a2200000 c 4500
001 1962438724
003 DE-627
005 20260224154046.0
007 cr uuu---uuuuu
008 260224s2025 xx |||||o 00| ||eng c
024 7 |a 10.1007/s10851-025-01239-9  |2 doi 
035 |a (DE-627)1962438724 
035 |a (DE-599)KXP1962438724 
040 |a DE-627  |b ger  |c DE-627  |e rda 
041 |a eng 
084 |a 27  |2 sdnb 
100 1 |a Boll, Bastian  |d 1996-  |e VerfasserIn  |0 (DE-588)1241416443  |0 (DE-627)1770926682  |4 aut 
245 1 0 |a Generative assignment flows for representing and learning joint distributions of discrete data  |c Bastian Boll, Daniel Gonzalez-Alvarado, Stefania Petra, Christoph Schnörr 
264 1 |c 28 May 2025 
300 |b Illustrationen 
300 |a 24 
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 24.02.2026 
520 |a We introduce a novel generative model for the representation of joint probability distributions of a possibly large number of discrete random variables. The approach uses measure transport by randomized assignment flows on the statistical submanifold of factorizing distributions, which enables to represent and sample efficiently from any target distribution and to assess the likelihood of unseen data points. The complexity of the target distribution only depends on the parametrization of the affinity function of the dynamical assignment flow system. Our model can be trained in a simulation-free manner by conditional Riemannian flow matching, using the training data encoded as geodesics on the assignment manifold in a closed form, with respect to the e-connection of information geometry. Numerical experiments devoted to distributions of structured image labelings demonstrate the applicability to large-scale problems, which may include discrete distributions in other application areas. Performance measures show that our approach scales better with the increasing number of classes than the recent related work. 
650 4 |a 49Q22 
650 4 |a 53B12 
650 4 |a 62H35 
650 4 |a 68T05 
650 4 |a 68U10 
650 4 |a 91A22 
650 4 |a Assignment flows 
650 4 |a Discrete distributions 
650 4 |a Generative models 
650 4 |a Information geometry 
650 4 |a Neural ODEs 
650 4 |a Normalizing flows 
650 4 |a Replicator equation 
700 1 |a Gonzalez Alvarado, Daniel  |e VerfasserIn  |0 (DE-588)1391173716  |0 (DE-627)1962440036  |4 aut 
700 1 |a Petra, Stefania  |e VerfasserIn  |0 (DE-588)1065905580  |0 (DE-627)816924961  |0 (DE-576)425560155  |4 aut 
700 1 |a Schnörr, Christoph  |e VerfasserIn  |0 (DE-588)1023033348  |0 (DE-627)717351017  |0 (DE-576)168404540  |4 aut 
773 0 8 |i Enthalten in  |t Journal of mathematical imaging and vision  |d Dordrecht [u.a.] : Springer Science + Business Media B.V, 1992  |g 67(2025), 3, Artikel-ID 34, Seite 1-24  |h Online-Ressource  |w (DE-627)271179465  |w (DE-600)1479363-5  |w (DE-576)110512847  |x 1573-7683  |7 nnas  |a Generative assignment flows for representing and learning joint distributions of discrete data 
773 1 8 |g volume:67  |g year:2025  |g number:3  |g elocationid:34  |g pages:1-24  |g extent:24  |a Generative assignment flows for representing and learning joint distributions of discrete data 
856 4 0 |u https://doi.org/10.1007/s10851-025-01239-9  |x Verlag  |x Resolving-System  |z kostenfrei  |3 Volltext  |7 0 
856 4 0 |u https://link.springer.com/article/10.1007/s10851-025-01239-9  |x Verlag  |z kostenfrei  |3 Volltext  |7 0 
951 |a AR 
992 |a 20260224 
993 |a Article 
994 |a 2025 
998 |g 1023033348  |a Schnörr, Christoph  |m 1023033348:Schnörr, Christoph  |d 110000  |d 110400  |e 110000PS1023033348  |e 110400PS1023033348  |k 0/110000/  |k 1/110000/110400/  |p 4  |y j 
998 |g 1065905580  |a Petra, Stefania  |m 1065905580:Petra, Stefania  |d 110000  |d 110200  |d 110000  |d 110400  |e 110000PP1065905580  |e 110200PP1065905580  |e 110000PP1065905580  |e 110400PP1065905580  |k 0/110000/  |k 1/110000/110200/  |k 0/110000/  |k 1/110000/110400/  |p 3 
998 |g 1391173716  |a Gonzalez-Alvarado, Daniel  |m 1391173716:Gonzalez-Alvarado, Daniel  |d 110000  |d 110400  |d 110000  |e 110000PG1391173716  |e 110400PG1391173716  |e 110000PG1391173716  |k 0/110000/  |k 1/110000/110400/  |k 0/110000/  |p 2 
998 |g 1241416443  |a Boll, Bastian  |m 1241416443:Boll, Bastian  |d 910000  |d 950000  |d 950950  |e 910000PB1241416443  |e 950000PB1241416443  |e 950950PB1241416443  |k 0/910000/  |k 1/910000/950000/  |k 2/910000/950000/950950/  |p 1  |x j 
999 |a KXP-PPN1962438724  |e 4921864845 
BIB |a Y 
SER |a journal 
JSO |a {"relHost":[{"physDesc":[{"extent":"Online-Ressource"}],"recId":"271179465","pubHistory":["1.1992 -"],"type":{"bibl":"periodical","media":"Online-Ressource"},"note":["Gesehen am 01.11.05"],"id":{"issn":["1573-7683"],"zdb":["1479363-5"],"eki":["271179465"]},"title":[{"title":"Journal of mathematical imaging and vision","title_sort":"Journal of mathematical imaging and vision"}],"language":["eng"],"origin":[{"publisherPlace":"Dordrecht [u.a.] ; Dordrecht [u.a.]","publisher":"Springer Science + Business Media B.V ; Kluwer","dateIssuedDisp":"1992-","dateIssuedKey":"1992"}],"disp":"Generative assignment flows for representing and learning joint distributions of discrete dataJournal of mathematical imaging and vision","part":{"volume":"67","text":"67(2025), 3, Artikel-ID 34, Seite 1-24","issue":"3","year":"2025","extent":"24","pages":"1-24"}}],"recId":"1962438724","language":["eng"],"note":["Gesehen am 24.02.2026"],"title":[{"title_sort":"Generative assignment flows for representing and learning joint distributions of discrete data","title":"Generative assignment flows for representing and learning joint distributions of discrete data"}],"person":[{"display":"Boll, Bastian","family":"Boll","role":"aut","given":"Bastian"},{"family":"Gonzalez Alvarado","display":"Gonzalez Alvarado, Daniel","given":"Daniel","role":"aut"},{"given":"Stefania","role":"aut","family":"Petra","display":"Petra, Stefania"},{"display":"Schnörr, Christoph","family":"Schnörr","role":"aut","given":"Christoph"}],"type":{"media":"Online-Ressource","bibl":"article-journal"},"physDesc":[{"noteIll":"Illustrationen","extent":"24 S."}],"id":{"eki":["1962438724"],"doi":["10.1007/s10851-025-01239-9"]},"name":{"displayForm":["Bastian Boll, Daniel Gonzalez-Alvarado, Stefania Petra, Christoph Schnörr"]},"origin":[{"dateIssuedDisp":"28 May 2025","dateIssuedKey":"2025"}]} 
SRT |a BOLLBASTIAGENERATIVE2820