Principled Interpolation in Normalizing Flows

Publikation: Beiträge in SammelwerkenAufsätze in KonferenzbändenForschungbegutachtet

Standard

Principled Interpolation in Normalizing Flows. / Fadel, Samuel; Mair, Sebastian; da Silva Torres, Ricardo et al.
Machine Learning and Knowledge Discovery in Databases. Research Track: European Conference, ECML PKDD 2021, Bilbao, Spain, September 13–17, 2021, Proceedings, Part II. Hrsg. / Nuria Oliver; Fernando Pérez-Cruz; Stefan Kramer; Jesse Read; Jose A. Lozano. Cham: Springer Nature, 2021. S. 116-131 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Band 12976 LNAI).

Publikation: Beiträge in SammelwerkenAufsätze in KonferenzbändenForschungbegutachtet

Harvard

Fadel, S, Mair, S, da Silva Torres, R & Brefeld, U 2021, Principled Interpolation in Normalizing Flows. in N Oliver, F Pérez-Cruz, S Kramer, J Read & JA Lozano (Hrsg.), Machine Learning and Knowledge Discovery in Databases. Research Track: European Conference, ECML PKDD 2021, Bilbao, Spain, September 13–17, 2021, Proceedings, Part II. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), Bd. 12976 LNAI, Springer Nature, Cham, S. 116-131, European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases - ECML PKDD 2021, Virtual, Online, 13.09.21. https://doi.org/10.1007/978-3-030-86520-7_8

APA

Fadel, S., Mair, S., da Silva Torres, R., & Brefeld, U. (2021). Principled Interpolation in Normalizing Flows. In N. Oliver, F. Pérez-Cruz, S. Kramer, J. Read, & J. A. Lozano (Hrsg.), Machine Learning and Knowledge Discovery in Databases. Research Track: European Conference, ECML PKDD 2021, Bilbao, Spain, September 13–17, 2021, Proceedings, Part II (S. 116-131). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Band 12976 LNAI). Springer Nature. https://doi.org/10.1007/978-3-030-86520-7_8

Vancouver

Fadel S, Mair S, da Silva Torres R, Brefeld U. Principled Interpolation in Normalizing Flows. in Oliver N, Pérez-Cruz F, Kramer S, Read J, Lozano JA, Hrsg., Machine Learning and Knowledge Discovery in Databases. Research Track: European Conference, ECML PKDD 2021, Bilbao, Spain, September 13–17, 2021, Proceedings, Part II. Cham: Springer Nature. 2021. S. 116-131. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). Epub 2021. doi: 10.1007/978-3-030-86520-7_8

Bibtex

@inbook{3266e4f6f901427a84649ba43db76f8f,
title = "Principled Interpolation in Normalizing Flows",
abstract = "Generative models based on normalizing flows are very successful in modeling complex data distributions using simpler ones. However, straightforward linear interpolations show unexpected side effects, as interpolation paths lie outside the area where samples are observed. This is caused by the standard choice of Gaussian base distributions and can be seen in the norms of the interpolated samples as they are outside the data manifold. This observation suggests that changing the way of interpolating should generally result in better interpolations, but it is not clear how to do that in an unambiguous way. In this paper, we solve this issue by enforcing a specific manifold and, hence, change the base distribution, to allow for a principled way of interpolation. Specifically, we use the Dirichlet and von Mises-Fisher base distributions on the probability simplex and the hypersphere, respectively. Our experimental results show superior performance in terms of bits per dimension, Frechet Inception Distance (FID), and Kernel Inception Distance (KID) scores for interpolation, while maintaining the generative performance",
keywords = "Informatics, Generative Modeling, Density estimation, Normalizing Flows, Business informatics",
author = "Samuel Fadel and Sebastian Mair and {da Silva Torres}, Ricardo and Ulf Brefeld",
note = "Funding Information: This research was financed in part by the Coordena??o de Aperfei?oamento de Pessoal de N?vel Superior (CAPES), Brazil, Finance Code 001 and by FAPESP (grants 2017/24005-2 and 2018/19350-5). Publisher Copyright: {\textcopyright} 2021, Springer Nature Switzerland AG.; European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases - ECML PKDD 2021, ECML PKDD ; Conference date: 13-09-2021 Through 17-09-2021",
year = "2021",
month = sep,
doi = "10.1007/978-3-030-86520-7_8",
language = "English",
isbn = "978-3-030-86519-1",
series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
publisher = "Springer Nature",
pages = "116--131",
editor = "Nuria Oliver and Fernando P{\'e}rez-Cruz and Stefan Kramer and Jesse Read and Lozano, {Jose A.}",
booktitle = "Machine Learning and Knowledge Discovery in Databases. Research Track",
address = "Germany",

}

RIS

TY - CHAP

T1 - Principled Interpolation in Normalizing Flows

AU - Fadel, Samuel

AU - Mair, Sebastian

AU - da Silva Torres, Ricardo

AU - Brefeld, Ulf

N1 - Funding Information: This research was financed in part by the Coordena??o de Aperfei?oamento de Pessoal de N?vel Superior (CAPES), Brazil, Finance Code 001 and by FAPESP (grants 2017/24005-2 and 2018/19350-5). Publisher Copyright: © 2021, Springer Nature Switzerland AG.

PY - 2021/9

Y1 - 2021/9

N2 - Generative models based on normalizing flows are very successful in modeling complex data distributions using simpler ones. However, straightforward linear interpolations show unexpected side effects, as interpolation paths lie outside the area where samples are observed. This is caused by the standard choice of Gaussian base distributions and can be seen in the norms of the interpolated samples as they are outside the data manifold. This observation suggests that changing the way of interpolating should generally result in better interpolations, but it is not clear how to do that in an unambiguous way. In this paper, we solve this issue by enforcing a specific manifold and, hence, change the base distribution, to allow for a principled way of interpolation. Specifically, we use the Dirichlet and von Mises-Fisher base distributions on the probability simplex and the hypersphere, respectively. Our experimental results show superior performance in terms of bits per dimension, Frechet Inception Distance (FID), and Kernel Inception Distance (KID) scores for interpolation, while maintaining the generative performance

AB - Generative models based on normalizing flows are very successful in modeling complex data distributions using simpler ones. However, straightforward linear interpolations show unexpected side effects, as interpolation paths lie outside the area where samples are observed. This is caused by the standard choice of Gaussian base distributions and can be seen in the norms of the interpolated samples as they are outside the data manifold. This observation suggests that changing the way of interpolating should generally result in better interpolations, but it is not clear how to do that in an unambiguous way. In this paper, we solve this issue by enforcing a specific manifold and, hence, change the base distribution, to allow for a principled way of interpolation. Specifically, we use the Dirichlet and von Mises-Fisher base distributions on the probability simplex and the hypersphere, respectively. Our experimental results show superior performance in terms of bits per dimension, Frechet Inception Distance (FID), and Kernel Inception Distance (KID) scores for interpolation, while maintaining the generative performance

KW - Informatics

KW - Generative Modeling

KW - Density estimation

KW - Normalizing Flows

KW - Business informatics

UR - http://www.scopus.com/inward/record.url?scp=85115734547&partnerID=8YFLogxK

UR - https://www.mendeley.com/catalogue/8bb702c5-63b3-3bd3-a647-8605a1e2eca2/

U2 - 10.1007/978-3-030-86520-7_8

DO - 10.1007/978-3-030-86520-7_8

M3 - Article in conference proceedings

AN - SCOPUS:85115734547

SN - 978-3-030-86519-1

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 116

EP - 131

BT - Machine Learning and Knowledge Discovery in Databases. Research Track

A2 - Oliver, Nuria

A2 - Pérez-Cruz, Fernando

A2 - Kramer, Stefan

A2 - Read, Jesse

A2 - Lozano, Jose A.

PB - Springer Nature

CY - Cham

T2 - European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases - ECML PKDD 2021

Y2 - 13 September 2021 through 17 September 2021

ER -

DOI

Zuletzt angesehen

Publikationen

  1. Social identity and place-based dynamics in community resilience building for natural disasters
  2. Parameters, concepts and the terminology of outer space law: a review of the essential facilities served by outer space activities and the rules of interpretation for treaty law and soft law guidelines.
  3. How to serve sustainability performance in businesses?
  4. Can radicals get a seat on the negotiation table? A Dynamic Perspective on Legitimation Processes.
  5. DeFacto - Temporal and multilingual deep fact validation
  6. Registered Replication Report on Srull and Wyer (1979)
  7. A Global Synthesis of Jatropha Cultivation
  8. Managing increasing environmental risks through agro-biodiversity and agri-environmental policies
  9. The effects of an active development of the mental model in the training process
  10. Recycling-oriented fabrication of soft robots
  11. Can Pulsed Electric Fields Treated Algal Cells Be Used as Stationary Phase in Chromatography?
  12. Unpacking the Personal Initiative-Performance Relationship
  13. The ten principles of green sample preparation
  14. Christian Steinbacher
  15. Calibration of the Chemcatcher ® passive sampler for monitoring selected polar and semi-polar pesticides in surface water
  16. Uncovering Divergence
  17. The continued relevance of compromissory clauses as a source of ICJ jurisdiction
  18. Exploring the Poincaré Ellipsis
  19. Taboo metaphtonymy, gender, and impoliteness
  20. The aftermath of colonization and forms of transactions between agriculture and industry - Exchange relations between industrialized and developing countries in the world economy of tobacco
  21. Managing uncertainty in the marketing of new-technology products
  22. Genau wahrnehmen - differenziert fördern
  23. Materialität der Diagramme
  24. Assessment of the biotic and abiotic elimination processes of five micropollutants during cultivation of the green microalgae Acutodesmus obliquus
  25. 'Creativity and Innovation' in the Nineteenth Century
  26. The relationship between working alliance and client outcomes in coaching