A0671
Title: Sequential decision-making in public health
Authors: Mengyan Zhang - University of Oxford (United Kingdom) [presenting]
Abstract: Sequential decision-making methods in machine learning have vital applications in public health. By adaptively making decisions, collecting samples, and learning models, the environment can be understood more efficiently with fewer data points. For instance, policymakers can allocate limited testing resources to maximize information about disease distribution. This decision-making process is modeled as an iterative node classification problem on an undirected, unweighted graph, where nodes represent locations and edges indicate the movement of infectious agents. Findings can aid in designing cost-effective surveillance policies for emerging and endemic pathogens, accelerating disease detection in resource-constrained settings. Moreover, the causal structure is adaptively learnt while optimizing targets through interventions. For example, to minimize HIV viral load by selecting different treatments, graph agnostic causal Bayesian optimization is proposed, an algorithm that actively discovers the causal structure to achieve optimal outcomes. Additionally, methods are developed to address imperfect feedback challenges in public health applications, including non-response bias in survey design and aggregated feedback. The aim is to enhance public health strategies by integrating advanced machine learning techniques, ultimately contributing to more effective and efficient disease control and prevention.
