Refining Deep Generative Models via Discriminator Gradient Flow

By Abdul Fatir Ansari, M. L. Ang, Harold Soh. 2020

53 citations recorded by the source corpus.

Deep generative modeling has seen impressive advances in recent years, to the point where it is now commonplace to see simulated samples (e.g., images) that closely resemble real-world data. However, generation quality is generally inconsistent for any given model and can vary dramatically between samples. We introduce Discriminator Gradient flow (DGflow), a new technique that improves generated samples via the gradient flow of entropy-regularized f-divergences between the real and the generated data distributions. The gradient flow takes the form of a non-linear Fokker-Plank equation, which can be easily simulated by sampling from the equivalent McKean-Vlasov process. By refining inferior samples, our technique avoids wasteful sample rejection used by previous methods (DRS&MH-GAN). Compared to existing works that focus on specific GAN variants, we show our refinement approach can be applied to GANs with vector-valued critics and even other deep generative models such as VAEs and Normalizing Flows. Empirical results on multiple synthetic, image, and text datasets demonstrate that DGflow leads to significant improvement in the quality of generated samples for a variety of generative models, outperforming the state-of-the-art Discriminator Optimal Transport (DOT) and Discriminator Driven Latent Sampling (DDLS) methods.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Statistical and Nonlinear Physics, Generative Models

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Arash Vahdat
2. Anima Anandkumar
3. Weili Nie
4. Patrick Hall
5. Kamyar Azizzadenesheli
6. Daniel J. Hsu
7. Rong Ge
8. Emma Brunskill

In Computer Vision and Pattern Recognition

1. Anima Anandkumar
2. Weili Nie
3. Arash Vahdat
4. Patrick Hall
5. Luc Van Gool
6. Jan Kautz
7. Ming-Hsuan Yang
8. Alan L. Yuille

In Statistical and Nonlinear Physics

1. Anima Anandkumar
2. Patrick Hall
3. Weili Nie
4. Arash Vahdat
5. Zong-Yi Li

In Generative Models

1. Anima Anandkumar
2. Weili Nie
3. Arash Vahdat
4. Karsten Kreis
5. Jan Kautz
6. Rewon Child

Citations in the corpus

• Score-based Generative Modeling in Latent Space (2021, neutral): “Our work also bears a resemblance to recent methods on improving the sampling quality in generative adversarial networks using gradient flows in the latent space [65, 66, 67, 68], with the main difference that these prior works use a discriminator to update the latent variables, whereas we train an ”
• Towards a Standard for Identifying and Managing Bias in Artificial Intelligence (2022, background): “Early practice has shown that concerns about the use of LLMs are indeed valid, with preliminary experimental results showing LLMs exhibit significant bias [158, 165, 166].”
• Controllable and Compositional Generation with Latent-Space Energy-Based Models (2021, methodology): “While [3] relied on the Euler-Maruyama solution of the SDE, we convert our generative SDE to an ODE.”