Initialization-Aware Score-Based Diffusion Sampling
This work addresses the high computational cost of diffusion sampling for practitioners, though it appears incremental as it builds on existing score-based frameworks.
The paper tackles the computational inefficiency of score-based diffusion models by analyzing how initialization affects convergence and proposing a method to learn optimal reverse-time initialization, achieving competitive or improved generative quality with significantly fewer sampling steps.
Score-based generative models (SGMs) aim at generating samples from a target distribution by approximating the reverse-time dynamics of a stochastic differential equation. Despite their strong empirical performance, classical samplers initialized from a Gaussian distribution require a long time horizon noising typically inducing a large number of discretization steps and high computational cost. In this work, we present a Kullback-Leibler convergence analysis of Variance Exploding diffusion samplers that highlights the critical role of the backward process initialization. Based on this result, we propose a theoretically grounded sampling strategy that learns the reverse-time initialization, directly minimizing the initialization error. The resulting procedure is independent of the specific score training procedure, network architecture, and discretization scheme. Experiments on toy distributions and benchmark datasets demonstrate competitive or improved generative quality while using significantly fewer sampling steps.