Combating Mode Collapse in GAN training: An Empirical Analysis using Hessian Eigenvalues
- Generative adversarial networks (GANs) provide state-of-the-art results in image generation. However, despite being so powerful, they still remain very challenging to train. This is in particular caused by their highly non-convex optimization space leading to a number of instabilities. Among them, mode collapse stands out as one of the most daunting ones. This undesirable event occurs when theGenerative adversarial networks (GANs) provide state-of-the-art results in image generation. However, despite being so powerful, they still remain very challenging to train. This is in particular caused by their highly non-convex optimization space leading to a number of instabilities. Among them, mode collapse stands out as one of the most daunting ones. This undesirable event occurs when the model can only fit a few modes of the data distribution, while ignoring the majority of them. In this work, we combat mode collapse using second-order gradient information. To do so, we analyse the loss surface through its Hessian eigenvalues, and show that mode collapse is related to the convergence towards sharp minima. In particular, we observe how the eigenvalues of the are directly correlated with the occurrence of mode collapse. Finally, motivated by these findings, we design a new optimization algorithm called nudged-Adam (NuGAN) that uses spectral information to overcome mode collapse, leading to empirically more stable convergence properties.…
Document Type: | Conference Proceeding |
---|---|
Conference Type: | Konferenzartikel |
Zitierlink: | https://opus.hs-offenburg.de/5289 | Bibliografische Angaben |
Title (English): | Combating Mode Collapse in GAN training: An Empirical Analysis using Hessian Eigenvalues |
Conference: | VISIGRAPP 2021: International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (16. : February 8-10, 2021 : Vienna, Austria) |
Author: | Ricard Durall Lopez, Avraam Chatzimichailidis, Peter Labus, Janis KeuperStaff MemberORCiDGND |
Year of Publication: | 2021 |
Publisher: | SciTePress |
Page Number: | 8 |
First Page: | 211 |
Last Page: | 218 |
Parent Title (English): | Proceedings of the 16th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications |
Editor: | Giovanni Maria Farinella, Petia Radeva, Jose Braz, Kadi Bouatouch |
Volume: | 4 |
ISBN: | 978-989-758-488-6 |
ISSN: | 2184-4321 |
DOI: | https://doi.org/10.5220/0010167902110218 |
URL: | https://www.scitepress.org/Papers/2021/101679/101679.pdf |
Language: | English | Inhaltliche Informationen |
Institutes: | Fakultät Elektrotechnik, Medizintechnik und Informatik (EMI) (ab 04/2019) |
Forschung / IMLA - Institute for Machine Learning and Analytics | |
Institutes: | Bibliografie |
Tag: | Eigenvalues; Generative Adversarial Network; Mode Collapse; Second-order Optimization; Stability | Formale Angaben |
Open Access: | Open Access |
Licence (German): | Creative Commons - CC BY-NC-ND - Namensnennung - Nicht kommerziell - Keine Bearbeitungen 4.0 International |