Differential-Critic GAN: Generating What You Want by a Cue of Preferences

Yinghua Yao, Yuangang Pan, Ivor Tsang, Xin Yao

Keywords: GAN, user-desired data distribution, user preference, critic

Abstract: This paper proposes Differential-Critic Generative Adversarial Network (DiCGAN) to learn the distribution of user-desired data when only partial instead of the entire dataset possesses the desired properties. Existing approaches select the desired samples first and train regular GANs on the selected samples to derive the user-desired data distribution. DiCGAN introduces a differential critic that can learn the preference direction from the pairwise preferences over the entire dataset. The resultant critic would guide the generation of the desired data instead of the whole data. Specifically, apart from the Wasserstein GAN loss, a ranking loss of the pairwise preferences is defined over the critic. It endows the difference of critic values between each pair of samples with the pairwise preference relation. The higher critic value indicates that the sample is preferred by the user. Thus training the generative model for higher critic values would encourage generating the user-preferred samples. Extensive experiments show that our DiCGAN can learn the user-desired data distributions.

One-sentence Summary: This paper proposes DiCGAN to learn the distribution of user-desired data from the entire dataset using pairwise preferences, where a differential critic is introduced to learn the preference direction from the pairwise preferences.

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics

Reviewed Version (pdf): https://openreview.net/references/pdf?id=-ftoYGPFz