STM-GAN: Sequentially Trained Multiple Generators for Mitigating Mode Collapse

Varshney, Sakshi and Srijith, P. K. and Balasubramanian, Vineeth N. (2020) STM-GAN: Sequentially Trained Multiple Generators for Mitigating Mode Collapse. In: 27th International Conference on Neural Information Processing, ICONIP 2020, 18-22 November 2020, Bangkok.

Full text not available from this repository. (Request a copy)


Generative adversarial networks have shown promise in generating images and videos. However, they suffer from the mode collapse issue which prevents it from generating complex multi-modal data. In this paper, We propose an approach to mitigate the mode collapse issue in generative adversarial networks (GANs). We propose to use multiple generators to capture various modes and each generator is encouraged to learn a different mode through a novel loss function. The generators are trained in a sequential way to effectively learn multiple modes. The effectiveness of the proposed approach is demonstrated through experiments on a synthetic data set, image data sets such as MNIST and fashion MNIST, and in multi-topic document modelling.

[error in script]
IITH Creators:
IITH CreatorsORCiD
Balasubramanian, Vineeth N
Srijith, P K
Item Type: Conference or Workshop Item (Paper)
Uncontrolled Keywords: Adversarial networks; Image datasets; Loss functions; Multi-modal data; Multi-topic; Multiple modes; Synthetic datasets
Subjects: Computer science
Divisions: Department of Computer Science & Engineering
Depositing User: . LibTrainee 2021
Date Deposited: 02 Aug 2021 04:02
Last Modified: 21 Nov 2022 04:35
Publisher URL:
OA policy:
Related URLs:

Actions (login required)

View Item View Item
Statistics for RAIITH ePrint 8618 Statistics for this ePrint Item