DisCo-Diff: Enhancing Continuous Diffusion Models with Discrete Latents

TL;DR

DisCo-Diff introduces discrete latent variables into diffusion models, simplifying the learning process and improving performance across various data synthesis tasks without relying on pre-trained networks.

Contribution

The paper presents a novel framework that integrates learnable discrete latents into diffusion models, enhancing their ability to model complex data distributions.

Findings

DisCo-Diff achieves state-of-the-art FID scores on ImageNet-64/128.

Discrete latents reduce the complexity of the diffusion process.

Model performance improves across toy data, image synthesis, and molecular docking.

Abstract

Diffusion models (DMs) have revolutionized generative learning. They utilize a diffusion process to encode data into a simple Gaussian distribution. However, encoding a complex, potentially multimodal data distribution into a single continuous Gaussian distribution arguably represents an unnecessarily challenging learning problem. We propose Discrete-Continuous Latent Variable Diffusion Models (DisCo-Diff) to simplify this task by introducing complementary discrete latent variables. We augment DMs with learnable discrete latents, inferred with an encoder, and train DM and encoder end-to-end. DisCo-Diff does not rely on pre-trained networks, making the framework universally applicable. The discrete latents significantly simplify learning the DM's complex noise-to-data mapping by reducing the curvature of the DM's generative ODE. An additional autoregressive transformer models the distribution of the discrete latents, a simple step because DisCo-Diff requires only few discrete variables with small codebooks. We validate DisCo-Diff on toy data, several image synthesis tasks as well as molecular docking, and find that introducing discrete latents consistently improves model performance. For example, DisCo-Diff achieves state-of-the-art FID scores on class-conditioned ImageNet-64/128 datasets with ODE sampler.