We propose a classifier guided diffusion model for Offline Multi-Objective Optimization. The guidance classifier is a preference model that is defined based on the notion of pareto-dominance.

We view offline optimization from the new lens of a distributional translation task which can be modeled with a generalized Brownian Bridge Diffusion process mapping between the low-value and high-value input distributions.

We propose new optimistic online-to-batch conversions for acceleration and universality.