TIOLI-GAN: A Bargaining for Efficiency

This paper reframes the generative adversarial net (GAN) by introducing Nash bargaining, which is optimized by the social planner to suggest the take-it-or-leaveit (TIOLI) mechanism to model generator-discriminator interactions. The proposed framework integrates the objectives of the discriminator and generator within a socially optimal bargaining structure, ensuring efficient and stable learning dynamics. Empirical results show improved stability and efficiency comparing to the standard GAN, strengthening the theoretical foundation of TIOLI-GAN