Diverse Sampling in Diffusion Models with Marginal Preserving Particle Guidance

2026-05-07

AI Summary

Diffusion models often struggle to simultaneously achieve high sample diversity and distributional fidelity during generation. This work proposes EDDY, a mechanism that enhances diversity by applying repulsive perturbations to particle trajectories through a divergence-free antisymmetric matrix field constructed via kernel methods, without altering the marginal distribution. Leveraging the symmetry of the Fokker–Planck equation, EDDY enables joint particle-level diversity guidance without requiring additional training. An efficient approximation strategy is introduced to scale the method to high-dimensional perceptual embeddings. Experimental results demonstrate that EDDY significantly outperforms state-of-the-art baselines on both synthetic distributions and text-to-image generation tasks, effectively improving sample diversity while preserving high fidelity.

Abstract

We present EDDY (Exact-marginal Diversification via Divergence-free dYnamics), a guidance mechanism for diffusion and flow matching models that promotes diversity among samples generated while maintaining quality. EDDY exploits symmetries of the Fokker-Planck equation, using drift perturbations that change particle trajectories while preserving the evolving marginal distribution. We instantiate this principle through kernel-based anti-symmetric pairwise matrix fields, constructed from the repulsive directions. The resulting divergence-free dynamics promote diversity at the joint particle level while preserving each particle's marginal distribution without any additional training. As computing the guidance can be computationally expensive in cases such as text-to-image generation with perceptual embeddings, we propose practical approximations as an effective and efficient solution. Experiments on synthetic distributions and text-to-image generation show that EDDY improves diversity while maintaining strong distributional fidelity compared to common baselines.

Problem

• diffusion models
• sample diversity
• marginal distribution
• flow matching
• generative modeling

Innovation

• divergence-free dynamics
• marginal-preserving guidance
• diffusion models
• sample diversity
• Fokker-Planck symmetry