Principled Interpolation in Normalizing Flows

Publikation: Beiträge in SammelwerkenAufsätze in KonferenzbändenForschungbegutachtet

Authors

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
OriginalspracheEnglisch
TitelMachine Learning and Knowledge Discovery in Databases. Research Track : European Conference, ECML PKDD 2021, Bilbao, Spain, September 13–17, 2021, Proceedings, Part II
HerausgeberNuria Oliver, Fernando Pérez-Cruz, Stefan Kramer, Jesse Read, Jose A. Lozano
Anzahl der Seiten16
ErscheinungsortCham
VerlagSpringer Nature
Erscheinungsdatum09.2021
Seiten116-131
ISBN (Print)978-3-030-86519-1
ISBN (elektronisch)978-3-030-86520-7
DOIs
PublikationsstatusErschienen - 09.2021
VeranstaltungEuropean Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases - ECML PKDD 2021 - Virtual, Online
Dauer: 13.09.202117.09.2021

DOI

Zuletzt angesehen

Publikationen

  1. Cross-Fertilizing Qualitative Perspectives on Effects of a Mindfulness-Based Intervention: An Empirical Comparison of Four Methodical Approaches
  2. Collaboration and Open Science Initiatives in Primate Research
  3. Working time flexibility and work-life balance
  4. Oxygen dependence in the photoreaction of the pesticide metamitron
  5. Cyberpunk
  6. Gutes Leben vor Ort
  7. Dispute and morality in the perception of societal risks: extending the psychometric model
  8. Will participation foster the successful implementation of water framework directive?
  9. Deformation and Anchoring of AA 2024-T3 rivets within thin printed circuit boards
  10. Species richness stabilizes productivity via asynchrony and drought-tolerance diversity in a large-scale tree biodiversity experiment
  11. Das Problem der Unbestimmtheit des Rechts
  12. Pesticide and metabolite fate, release and transport modelling at catchment scale
  13. Forest Ecosystems: A functional and biodiversity perspective
  14. A PD Fuzzy Control of a Nonholonomic Car-Like Robot for Drive Assistant Systems
  15. Efficacy of an internet-based self-help intervention to reduce co-occurring alcohol misuse and depression symptoms in adults
  16. Predictive Maintenance of Bearings Through IoT and Cloud-Based Systems
  17. Ankunft einer Katze
  18. Orientierung im Realraum
  19. Grand theories and mid-range theories
  20. Does outcome expectancy predict outcomes in online depression prevention? Secondary analysis of randomised-controlled trials
  21. Forecasting Government Bond Yields with Neural Networks Considering Cointegration
  22. CULTURAL TENSIONS AND VALUES-ACTION GAPS IN SUSTAINABILITY-ORIENTED INNOVATION
  23. Using the Rapid Impact Assessment Matrix to synthesize biofuel and bioenergy impact assessment results