Abstract:Diffusion models suffer from slow sample generation at inference time. Despite recent efforts, improving the sampling efficiency of stochastic samplers for diffusion models remains a promising direction. We propose Splitting Integrators for fast stochastic sampling in pre-trained diffusion models in augmented spaces. Commonly used in molecular dynamics, splitting-based integrators attempt to improve sampling efficiency by cleverly alternating between numerical updates involving the data, auxiliary, or noise variables. However, we show that a naive application of splitting integrators is sub-optimal for fast sampling. Consequently, we propose several principled modifications to naive splitting samplers for improving sampling efficiency and denote the resulting samplers as Reduced Splitting Integrators. In the context of Phase Space Langevin Diffusion (PSLD) [Pandey \& Mandt, 2023] on CIFAR-10, our stochastic sampler achieves an FID score of 2.36 in only 100 network function evaluations (NFE) as compared to 2.63 for the best baselines.

What problem does this paper attempt to address?

### The Problem Addressed by the Paper This paper aims to address the issue of slow sample generation speed in diffusion models during the inference stage. Specifically, the authors propose the following points: 1. **Fast Random Sampling**: - Most current research on accelerating diffusion models focuses on deterministic samplers, but these methods cannot achieve optimal sample quality even under a large number of network function evaluations (NFE). - Random samplers usually achieve better sample quality but require more sampling budget than deterministic samplers. 2. **Enhancing Efficient Sampling of Diffusion Models**: - Most efforts to improve diffusion model sampling have focused on diffusion models in the data space. - Recent studies have shown that performing diffusion in the joint space (combining data space with auxiliary variables) can improve sample quality and likelihood estimation. - However, efficient sampling for enhanced diffusion models is still under-researched. ### Main Contributions - **Splitting Integrator**: - Inspired by molecular dynamics, the authors propose a method called "splitting integrator" to accelerate random sampling of pre-trained diffusion models. - The splitting integrator improves sampling efficiency by cleverly alternating updates of data, auxiliary, or noise variables. - The authors found that simple applications of the splitting integrator are not always optimal and proposed a series of improvements to further enhance sampling efficiency. - **Improved Splitting Integrator**: - Several improvements to the simple splitting integrator are proposed, named "Reduced Splitting Integrators." - These improvements include reusing score function evaluations, adjusting the amount of injected noise, etc. ### Experimental Results - On the CIFAR-10 and CelebA-64 datasets, the proposed splitting integrator sampler achieved significant performance improvements with fewer NFEs. - Specifically, with only 100 network function evaluations, the proposed random sampler achieved an FID score of 2.36 on CIFAR-10. In summary, this paper addresses the issue of fast random sampling in diffusion models by improving the splitting integrator method and validates its effectiveness through experiments.

Towards Fast Stochastic Sampling in Diffusion Generative Models

Efficient Integrators for Diffusion Generative Models

Fast Diffusion Probabilistic Model Sampling through the lens of Backward Error Analysis

Accelerating Diffusion Models with Parallel Sampling: Inference at Sub-Linear Time Complexity

Towards Faster Non-Asymptotic Convergence for Diffusion-Based Generative Models

On Accelerating Diffusion-Based Sampling Process via Improved Integration Approximation

Plug-and-Play Split Gibbs Sampler: Embedding Deep Generative Priors in Bayesian Inference

Simple and Fast Distillation of Diffusion Models

On diffusion models for amortized inference: Benchmarking and improving stochastic control and sampling

Fast constrained sampling in pre-trained diffusion models

Accelerating Parallel Sampling of Diffusion Models

Fast Sampling via Discrete Non-Markov Diffusion Models with Predetermined Transition Time

New algorithms for sampling and diffusion models

Accelerating Convergence of Score-Based Diffusion Models, Provably

Exploring Guided Sampling of Conditional GANs

Stochastic Runge-Kutta Methods: Provable Acceleration of Diffusion Models

Faster Diffusion Sampling with Randomized Midpoints: Sequential and Parallel

Diffusion Sampling Correction via Approximately 10 Parameters

A Practical Diffusion Path for Sampling

Diffusion Generative Flow Samplers: Improving learning signals through partial trajectory optimization

Implicit Diffusion: Efficient Optimization through Stochastic Sampling