ParetoPilot: Zero-Surrogate Offline Multi-Objective Optimization via Infer-Perturb-Guide Diffusion
Researchers have introduced ParetoPilot, a new framework for offline multi-objective optimization that eliminates the need for external surrogate models. This zero-surrogate diffusion approach leverages pre-trained diffusion models by incorporating an Infer-Perturb-Guide engine. This engine infers objective directions and applies forces for convergence and diversity, guiding the generation process. Experiments show ParetoPilot outperforms existing surrogate-based methods across numerous tasks, offering improved Pareto front coverage and data privacy. AI
IMPACT Introduces a novel method for optimizing designs using diffusion models, potentially improving efficiency and privacy in generative design tasks.