Training GANs with centripetal acceleration. (2nd September 2020)

• Record Type: Journal Article
• Authors: Peng, Wei; Dai, Yu-Hong; Zhang, Hui; Cheng, Lizhi
• Is Part Of: Optimization methods and software. Volume 35:Number 5(2020)
• Page Start: 955
• Page End: 973
• Publication Date: 2020-09-02
• Subjects: Primary: 97N60 -- 90C25 -- Secondary: 90C06 -- 90C30; Mathematical optimization -- Periodicals; Algorithms -- Periodicals; 519.7
• DOI: 10.1080/10556788.2020.1754414
• Languages: English
• ISSNs: 1055-6788

Abstract

ABSTRACT: Training generative adversarial networks (GANs) often suffers from cyclic behaviours of iterates. Based on a simple intuition that the direction of centripetal acceleration of an object moving in uniform circular motion is toward the centre of the circle, we present the Simultaneous Centripetal Acceleration (SCA) method and the Alternating Centripetal Acceleration (ACA) method to alleviate the cyclic behaviours. Under suitable conditions, gradient descent methods with either SCA or ACA are shown to be linearly convergent for bilinear games. Numerical experiments are conducted by applying ACA to existing gradient-based algorithms in a GAN setup scenario, which demonstrate the superiority of ACA.