Bidirectional Generative Modeling Using Adversarial Gradient Estimation
This work provides a general recipe for principled f-divergence-based generative modeling, which is incremental as it unifies and extends existing methods like VAE and BiGAN.
The paper tackles the problem of bidirectional generative modeling under a general f-divergence formulation, presenting a new optimization method using adversarial gradient estimation, and demonstrates its advantage over existing methods through theoretical and empirical studies.
This paper considers the general $f$-divergence formulation of bidirectional generative modeling, which includes VAE and BiGAN as special cases. We present a new optimization method for this formulation, where the gradient is computed using an adversarially learned discriminator. In our framework, we show that different divergences induce similar algorithms in terms of gradient evaluation, except with different scaling. Therefore this paper gives a general recipe for a class of principled $f$-divergence based generative modeling methods. Theoretical justifications and extensive empirical studies are provided to demonstrate the advantage of our approach over existing methods.