Diffusion Crossover: Defining Evolutionary Recombination in Diffusion Models via Noise Sequence Interpolation

2026-04-16

AI Summary

This work addresses the limitation of interactive evolutionary computation in high-dimensional generative spaces, where the absence of semantically consistent crossover operations leads to overreliance on mutation. For the first time, it explicitly defines a crossover operator within diffusion models by fusing parental noise trajectories via spherical linear interpolation (Slerp), enabling structure-preserving recombination during the denoising process. The method introduces a timestep-controllable interpolation strategy that flexibly balances exploration and exploitation. Experimental results demonstrate that offspring images exhibit perceptual smoothness and semantic consistency as measured by PCA and LPIPS metrics. Human-in-the-loop evaluations further confirm that the approach effectively supports user-guided image exploration, endowing diffusion models with a novel capability for structured evolutionary search.

Technology Category

Computer Vision: Diffusion Models for VisionSearch and Optimization: Evolutionary ComputationMachine Learning: Evolutionary Learning

Application Category

Search and Retrieval-Augmented AI: Large language models for searchUser Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalizationSemantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

Interactive Evolutionary Computation (IEC) provides a powerful framework for optimizing subjective criteria such as human preferences and aesthetics, yet it suffers from a fundamental limitation: in high-dimensional generative representations, defining crossover in a semantically consistent manner is difficult, often leading to a mutation-dominated search. In this work, we explicitly define crossover in diffusion models. We propose Diffusion crossover, which formulates evolutionary recombination as step-wise interpolation of noise sequences in the reverse process of Denoising Diffusion Probabilistic Models (DDPMs). By applying spherical linear interpolation (Slerp) to the noise sequences associated with selected parent images, the proposed method generates offspring that inherit characteristics from both parents while preserving the geometric structure of the diffusion process. Furthermore, controlling the time-step range of interpolation enables a principled trade-off between diversity (exploration) and convergence (exploitation). Experimental results using PCA analysis and perceptual similarity metrics (LPIPS) demonstrate that Diffusion crossover produces perceptually smooth and semantically consistent transitions between parent images. Qualitative interactive evolution experiments further confirm that the proposed method effectively supports human-in-the-loop image exploration. These findings suggest a new perspective: diffusion models are not only powerful generators, but also structured evolutionary search spaces in which recombination can be explicitly defined and controlled.

Problem

• Interactive Evolutionary Computation
• crossover
• diffusion models
• semantic consistency
• high-dimensional generative representations

Innovation

• Diffusion Crossover
• Noise Sequence Interpolation
• Evolutionary Recombination
• Denoising Diffusion Probabilistic Models
• Interactive Evolutionary Computation