Optimizing for the Shortest Path in Denoising Diffusion Model

2025-03-05

AI Summary

This paper addresses the low residual propagation efficiency and limited generation quality in denoising diffusion models. We propose ShortDF, the first diffusion model that formulates the reverse denoising process as an optimal path planning problem on a weighted directed graph, leveraging graph-theoretic shortest-path modeling. ShortDF enables global residual propagation through differentiable graph optimization and jointly learns both the initial residual and edge weights. Integrated within the DDIM framework, it supports end-to-end training. On multiple benchmark datasets, ShortDF reduces sampling steps by up to 56.5% (2.3× speedup) while improving FID by 12.6%, significantly enhancing both generation speed and fidelity. These results empirically validate the effectiveness of incorporating graph-structured priors into diffusion dynamics modeling.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Search and Optimization: Learning to Search
• Machine Learning: Graph-based Machine Learning

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

In this research, we propose a novel denoising diffusion model based on shortest-path modeling that optimizes residual propagation to enhance both denoising efficiency and quality.Drawing on Denoising Diffusion Implicit Models (DDIM) and insights from graph theory, our model, termed the Shortest Path Diffusion Model (ShortDF), treats the denoising process as a shortest-path problem aimed at minimizing reconstruction error. By optimizing the initial residuals, we improve the efficiency of the reverse diffusion process and the quality of the generated samples.Extensive experiments on multiple standard benchmarks demonstrate that ShortDF significantly reduces diffusion time (or steps) while enhancing the visual fidelity of generated samples compared to prior arts.This work, we suppose, paves the way for interactive diffusion-based applications and establishes a foundation for rapid data generation. Code is available at https://github.com/UnicomAI/ShortDF.

Problem

• Optimizes residual propagation for denoising efficiency and quality.
• Treats denoising as a shortest-path problem to minimize reconstruction error.
• Reduces diffusion time while enhancing visual fidelity of generated samples.

Innovation

• Shortest-path modeling optimizes residual propagation
• Reduces diffusion steps, enhances sample quality
• Integrates DDIM with graph theory insights