Training Diffusion Language Models for Black-Box Optimization

TL;DR

This paper introduces a diffusion language model approach for offline black-box optimization, effectively capturing bidirectional dependencies and improving design discovery in limited-data scenarios.

Contribution

It adapts diffusion LLMs for BBO by constructing a unified prompt-response corpus and a two-stage post-training framework, addressing domain gaps and enhancing optimization performance.

Findings

Achieves state-of-the-art results on Design-Bench small-data settings.

Effectively models bidirectional dependencies in design problems.

Improves design optimization with limited labeled data.

Abstract

We study offline black-box optimization (BBO), aiming to discover improved designs from an offline dataset of designs and labels, a problem common in robotics, DNA, and materials science with limited labeled samples. While recent work applies autoregressive LLMs to BBO by formatting tasks as natural-language prompts, their left-to-right design generation struggles to capture the strong bidirectional dependencies inherent in design problems. To address this, we propose adapting diffusion LLMs to offline BBO to leverage their bidirectional modeling capabilities. However, a domain gap exists between the natural text pre-training of diffusion LLMs and the heterogeneous signals in BBO (prompts, designs, and labels). To bridge this gap, we construct a unified prompt-response corpus and introduce delimiter tokens to explicitly mark field boundaries for domain adaptation. We further propose a two-stage post-training framework to align the diffusion LLM generation with high-label designs. The first stage performs supervised fine-tuning on the unified dataset via masked-response prediction, and the second stage adopts reinforcement learning with rewards defined by label improvements. Our method achieves state-of-the-art results on Design-Bench small-data settings.