HingeRLC-GAN: Combating Mode Collapse with Hinge Loss and RLC Regularization
Osman Goni, Himadri Saha Arka, Mithun Halder, Mir Moynuddin Ahmed Shibly, Swakkhar Shatabda · 2025
Read HingeRLC-GAN: Combating Mode Collapse with Hinge Loss and RLC Regularization by Osman Goni, Himadri Saha Arka, Mithun Halder, Mir Moynuddin Ahmed Shibly, Swakkhar Shatabda (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-19074.txt.
Description
Osman Goni and co-authors present a 2025 paper on a chronic failure of image-generating networks. In mode collapse, a GAN's generator finds a few safe outputs and repeats them, ignoring the data's variety. The team attacks the problem on several fronts at once.
A Wasserstein loss with gradient penalty steadied training and widened coverage. Architecture trials singled out residual networks as diversity boosters. Their final model, HingeRLC-GAN, pairs a hinge loss with a new RLC regularizer. Scores on standard image quality measures, an FID of 18 and a KID of 0.001, beat the compared methods. The authors read the result as a better balance between training stability and output range. Diverse fakes matter wherever synthetic data trains other systems.
Catalog record
| CREATOR | Osman Goni, Himadri Saha Arka, Mithun Halder, Mir Moynuddin Ahmed Shibly, Swakkhar Shatabda |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 26 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2503-19074