Abstract

As public health challenges continue to grow, there are many opportunities for impactful computational research inspired by these problems and tested in real world settings. This session will explore the intersections between different research areas (anomaly detection, forecasting, resource optimization, and deployment) to meet practical use cases in public health.This tutorial will provide both newcomers and experienced researchers with essential strategies for impactful research and decision-making in critical, changing, and resource-constrained fields like public health.

Tutorial Description

Public health problems are highly impactful and include topics like infectious diseases, antimicrobial and drug abuse, diabetes, and mental health. However, the diversity, multi-modality, and privacy of health data present unique challenges in collecting and analyzing data. On the other hand, these challenges also present new opportunities to develop new machine learning and data mining techniques to handle complex health data, pushing the demand for reliable data collection and infrastructure, epidemiological modeling and analysis of other social determinants, causal inference and counterfactual analysis, and complex decision making. This tutorial comprehensively summarizes the challenges, opportunities, and experiences in designing data-driven decision-making solutions for public health applications.

Computer scientists need an interdisciplinary approach to better understand public health challenges, constraints, and the stakeholders involved. Towards that goal, this tutorial highlights the shift toward data-driven decision-making in public health, exploring how robust 1) data systems feed into 2) models that inform 3) resource optimization in constrained settings that are designed for 4) deployment. Along the way, we examine key challenges from a computational perspective, identifying interdisciplinary opportunities for impactful research.

In this tutorial, we will cover 4 different computational challenges relevant to public health.  

1.  Public Health Data Systems: Understanding data collection processes, including privacy, ambient monitoring, and tiered public health indicators, informs the challenges associated with modeling and decision-making. This is a rich area of innovation, include improving infrastructure (e.g., leveraging IoT sensors), addressing privacy via statistical approaches (e.g., federated learning), and developing multimodal methods to provide domain experts with contextual insights.

2. Public Health Modeling:  Building on these data insights, we share how to handle these data issues and dynamics into forecasting models, with a focus on designing differential simulations for epidemiology that can be seamlessly integrated into recent end-to-end optimization frameworks. We will share insights for a new class of 'hybrid models' that combine the domain knowledge of mechanistic epidemiological models with machine learning. These models integrate the data-driven strengths of AI/ML with traditional methods like ordinary differential equations.

3. Intervention Optimization: These optimization techniques are form a bridge to data-driven decision making. Intervention design and optimization is largely dependent on public health data where the model is learned from, as well as the deployment experience where the real-world constraints and domain knowledge are extracted. These questions, inspired by real-world challenges and constraints, motivate interesting research challenges in online learning and multi-armed bandits.

4. Deployment: Tying it all together, deploying such AI-powered models in the public health domain is a complex journey with numerous challenges that require careful navigation and constant adaptation. Through case studies, we will cover iterative model development, real-world complexities, responsible deployment so that advances in computing can be realized.
   
   

Slides

Organizers