Mitigating Mode Collapse using Multiple GANs Training System
0
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
Generative Adversarial Networks (GANs) are typically described as a two-player game between a generator and a discriminator, wherethe generator aims to produce realistic data, and the discriminator tries to distinguish between real and generated data. However, thissetup often leads to mode collapse, where the generator produces limited variations in the data, failing to capture the full range of thetarget data distribution. This paper proposes a new training system to mitigate the mode collapse problem. Specifically, it extends thetraditional two-player game of GANs into a multi-player game and introduces a peer-evaluation method to effectively train multiple GANs. In the peer-evaluation process, the generated samples from each GANs are evaluated by the other players. This provides external feedback,serving as an additional standard that helps GANs recognize mode failure. This cooperative yet competitive training method encouragesthe generators to explore and capture a broader range of the data distribution, mitigating mode collapse problem. This paper explainsthe detailed algorithm for peer-evaluation based multi-GANs training and validates the performance through experiments.
Abstract
Generative Adversarial Networks (GANs) are typically described as a two-player game between a generator and a discriminator, wherethe generator aims to produce realistic data, and the discriminator tries to distinguish between real and generated data. However, thissetup often leads to mode collapse, where the generator produces limited variations in the data, failing to capture the full range of thetarget data distribution. This paper proposes a new training system to mitigate the mode collapse problem. Specifically, it extends thetraditional two-player game of GANs into a multi-player game and introduces a peer-evaluation method to effectively train multiple GANs. In the peer-evaluation process, the generated samples from each GANs are evaluated by the other players. This provides external feedback,serving as an additional standard that helps GANs recognize mode failure. This cooperative yet competitive training method encouragesthe generators to explore and capture a broader range of the data distribution, mitigating mode collapse problem. This paper explainsthe detailed algorithm for peer-evaluation based multi-GANs training and validates the performance through experiments.
참고문헌 (20)
- A. Ramesh, "Zero-shot text-to-image generation", Proceedings of the Advances in Neural Information Processing Systems, pp.8821~8831 (2021)
- M. Tao, "Galip : Generative adversarial clips for text-to-image synthesis", pp.2023, Proceedings of the IEEE
- H. Alqahtani, "Applications of generative adversarial networks(gans) : An updated review", Archives of Computational Methods in Engineering, 28, pp.525~552 (2021)
- Z. Xiao, "Tackling the generative learning trilemma with denoising diffusion gans", Proceedings of the International Conference on Learning Representations (2022)
- I. Goodfellow, "Generative adversarial nets", Proceedings of the Advances in Neural Information Processing Systems, 27, pp.2672~2680 (2014)
- T. Salimans, "Improved techniques for training gans", Proceedings of the Advances in Neural Information Processing Systems, 29 (2016)
- J. Y. Shim, "A study on auction-inspired multi-gan training", Annual Spring Conference of KIPS, 30(1), pp.527~529 (2023)
- J. Y. Shim, "Enhancements and applications of gans:Cross-modal generation, captcha system and mode collapse problem", Ph.D, Department of Electrical and Computer Engineering, pp.2024
- M. Arjovsky, "Wasserstein generative adversarial networks", Proceedings of the International Conference on Machine Learning, pp.214~223 (2017)
- I. Gulrajani, "Improved training of wasserstein gans", Proceedings of the Advances in Neural Information Processing Systems, 30, pp.5767~5777 (2017)
- X. Mao, "Least squares generative adversarial networks", Proceedings of the IEEE International Conference on Computer Vision, pp.2794~2802 (2017)
- X. Chen, "Infogan: Interpretable representation learning by information maximizing generative adversarial nets", Proceedings of the Advances in Neural Information Processing Systems, 29 (2016)
- T. Che, "Mode regularized generative adversarial networks", Proceedings of the International Conference on Learning Representations (2017)
- M. Mohebbi Moghaddam, "Games of gans : gametheoretical models for generative adversarial networks", Artificial Intelligence Review, pp.1~37 (2023)
- A. Ghosh, "Multi-agent diverse generative adverarial networks", Proceedings of the IEEE, 2018, 8513~8521
- Q. Hoang, "MGAN : Training generative adversarial nets with multiple generators", International Conference on Learning Representations (2018)
- T. Nguyen, "Dual discriminator generative adversarial nets", Proceedings of the Advances in Neural Information Processing Systems, 30 (2017)
- J. Choi, "Mcl-gan : Generative adversarial networks with multiple specialized discriminators", Proceedings of the Advances in Neural Information Processing Systems, 35, pp.29 597~29 609 (2022)
- I. Albuquerque, "Multi-objective training of generative adversarial networks with multiple discriminators", Proceedings of the Advances in Neural Information Processing Systems, ser. Proceedings of Machine Learning Research, pp.202~211 (2019), PMLR