Policy-Embedded Graph Expansion: Networked HIV Testing with Diffusion-Driven Network Samples

This paper presents a novel framework for improving network-based HIV testing using AI-driven sequential decision-making. The authors introduce Policy-Embedded Graph Expansion (PEGE), which embeds a generative graph expansion process directly into testing policies, and Dynamics-Driven Branching (DDB), a diffusion-based graph model designed for data-limited referral networks. Developed in collaboration with the World Health Organization and Wits University, the approach improves HIV detection efficiency on real-world transmission networks, outperforming existing baselines while operating under realistic constraints relevant to public health deployment.