How does Lipschitz regularization influence GAN training?
- Submitting institution
-
Cardiff University / Prifysgol Caerdydd
- Unit of assessment
- 11 - Computer Science and Informatics
- Output identifier
- 113087258
- Type
- E - Conference contribution
- DOI
-
10.1007/978-3-030-58517-4_19
- Title of conference / published proceedings
- Lecture Notes in Computer Science
- First page
- 310
- Volume
- 0
- Issue
- -
- ISSN
- 0302-9743
- Open access status
- Compliant
- Month of publication
- October
- Year of publication
- 2020
- URL
-
https://doi.org/10.1007/978-3-030-58517-4_19
- Supplementary information
-
-
- Request cross-referral to
- -
- Output has been delayed by COVID-19
- No
- COVID-19 affected output statement
- -
- Forensic science
- No
- Criminology
- No
- Interdisciplinary
- No
- Number of additional authors
-
2
- Research group(s)
-
V - Visual computing
- Citation count
- -
- Proposed double-weighted
- No
- Reserve for an output with double weighting
- No
- Additional information
- This paper, resulting from collaboration with UCL/Adobe Research and KAUST, saved massive time and efforts in GAN loss function research by clarifying an important false belief in GAN training. The preprint of this paper was tweeted by Ian Goodfellow (Inventor of GAN, Director of Machine Learning at Apple Inc.) and requested to be included in one of his GAN overview presentations (https://twitter.com/goodfellow_ian/status/1128710626421891073?lang=en). A by-product of this research inspired an EPSRC DTP studentship titled “Uncovering the ‘Instincts’ of Deep Generative Models for Fair and Unbiased Visual Content Creation”. This paper was presented at ECCV 2020, and includes 62-page supplementary material (https://static-content.springer.com/esm/chp:10.1007/978-3-030-58517-4_19/MediaObjects/504471_1_En_19_MOESM1_ESM.pdf).
- Author contribution statement
- -
- Non-English
- No
- English abstract
- -