Robust Guided Diffusion for Offline Black-Box Optimization

2024-10-01

arXiv.org

AI Summary

We address offline black-box optimization—maximizing an unknown objective function using only a static dataset of design-performance pairs. We propose a two-module framework: surrogate-augmented sampling and diffusion-driven surrogate refinement. To our knowledge, this is the first method to synergistically integrate classifier-free diffusion models with surrogate-guided closed-loop optimization. First, gradient-guided sampling leverages regression surrogates (e.g., MLP or GBDT) to provide explicit gradient directions for diffusion-based design generation. Second, the surrogate is dynamically retrained using diffusion-generated samples as prior knowledge, thereby correcting its out-of-distribution prediction bias. Our approach achieves state-of-the-art performance across all benchmarks in Design-Bench, significantly improving both high-value sample generation rates and generalization to unseen design regions. The implementation is publicly available.

Technology Category

• Search and Optimization: Sampling/Simulation-based Search
• Computer Vision: Diffusion Models for Vision
• Machine Learning: Optimization

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Offline black-box optimization aims to maximize a black-box function using an offline dataset of designs and their measured properties. Two main approaches have emerged: the forward approach, which learns a mapping from input to its value, thereby acting as a proxy to guide optimization, and the inverse approach, which learns a mapping from value to input for conditional generation. (a) Although proxy-free~(classifier-free) diffusion shows promise in robustly modeling the inverse mapping, it lacks explicit guidance from proxies, essential for generating high-performance samples beyond the training distribution. Therefore, we propose extit{proxy-enhanced sampling} which utilizes the explicit guidance from a trained proxy to bolster proxy-free diffusion with enhanced sampling control. (b) Yet, the trained proxy is susceptible to out-of-distribution issues. To address this, we devise the module extit{diffusion-based proxy refinement}, which seamlessly integrates insights from proxy-free diffusion back into the proxy for refinement. To sum up, we propose extit{ extbf{R}obust extbf{G}uided extbf{D}iffusion for Offline Black-box Optimization}~( extbf{RGD}), combining the advantages of proxy~(explicit guidance) and proxy-free diffusion~(robustness) for effective conditional generation. RGD achieves state-of-the-art results on various design-bench tasks, underscoring its efficacy. Our code is at https://github.com/GGchen1997/RGD.

Problem

• Offline Blackbox Optimization
• Surrogate Model Guidance
• Unknown Design Handling

Innovation

• Robust Guided Diffusion
• Offline Black-box Optimization
• Enhanced Sampling and Refinement