ICDM 2025 Tutorial — AI‑Driven Multimodal Frameworks for Healthcare Decision‑Making

Abstract

Modern healthcare systems generate vast amounts of multimodal data, including electronic health records, clinical notes, medical images, physiological signals, and genomic sequences. While machine learning and data mining have shown remarkable promise, effectively unifying multimodal data into trustworthy, interpretable, and actionable frameworks remains a central challenge. This tutorial will provide a comprehensive overview of recent advances in AI-driven multimodal frameworks for healthcare, with four core themes: (1) foundation models for multimodal healthcare data, (2) neurosymbolic approaches for trustworthy high-stakes decisions, (3) integration of mechanistic modeling into machine learning for enhanced explainability, and (4) generative AI for population health applications. Through case studies and open discussions, we will highlight practical methodologies, key challenges, and future research directions. The tutorial is designed to bridge cutting-edge data mining research with pressing healthcare needs, fostering collaboration between AI researchers and healthcare practitioners.

Agenda & Contents

Introduction
Part 1 — Multi‑Modal Foundation Models for Healthcare

Healthcare data spans diverse modalities, including text, images, graphs, and time series, necessitating unified modeling for effective decision-making. While foundation models excel in single modalities, integrating heterogeneous signals remains difficult. Systems like Med-PaLM-2 show promise yet struggle with autonomy, personalization, and input sensitivity. This part reviews progress on multimodal fusion and presents our work mitigating multimodal bias and enhancing multi-agent LLM collaboration for clinical decision-making.
Part 2 — Neurosymbolic Learning for Trustworthy High‑Stakes Decision‑Making

High-stakes decisions demand AI systems that are accurate, interpretable, and reliable. This part examines how neuro-symbolic learning unites symbolic reasoning with neural networks to enhance explainability and uncertainty handling. By embedding logic-based knowledge into deep learning, these models better incorporate domain expertise and reason about constraints. We will review key techniques and highlight use cases demonstrating the power and trustworthiness of neurosymbolic approaches.
Break
Part 3 — Trustworthy ML with Mechanistic Modeling

Machine learning excels in integrating multimodal healthcare data and generating patient-specific predictions, yet struggles to incorporate explicit medical knowledge for explainability. Knowledge graphs help embed domain knowledge but lack mechanistic rigor. This part explores integrating mechanistic modeling into ML frameworks in closed form, enhancing prediction reliability, adherence to biological principles, and interpretability—thereby improving the trustworthiness of AI systems for clinical decision-making.
Part 4 — Generative AI for Population Health

Multimodal ML enables data-driven public health decisions by integrating diverse sources like social media, genomics, policy, and mobility data. Generative AI further advances this by synthesizing and simulating across modalities to support adaptive, human-centered strategies. This part reviews progress and challenges in generative AI for pandemic response, outbreak detection, and behavioral modeling, highlighting case studies and future directions for trustworthy, impactful public health applications.
Open Discussion
Closing

Target Audience

We aim to engage machine learning and data mining researchers from both academia and industry. This includes those developing foundational methods and seeking impactful applications, as well as applied scientists focused on deploying AI systems in practice. We also seek participation from healthcare professionals or data scientists who are seeking collaboration with the AI research community. It is recommended that the audience have a foundational knowledge of supervised/unsupervised learning, neural networks, and basic healthcare concepts. Furthermore, prior experience in LLMs will be beneficial for attendees.

Presenters

Jiaming Cui

Jiaming Cui is an Assistant Professor in the Department of Computer Science at Virginia Tech. His research aims to bridge public health with clinical decisions using machine learning and mechanistic modeling. He has published in leading science journals and top CS venues such as PNAS, NPJ Digital Medicine, NeurIPS, ICML, AAAI, UAI, and SDM. He has closely collaborated with clinicians, and his work has been applied in multiple healthcare facilities. His work has also significantly contributed to pandemic prediction and prevention in the past several years, including helping decision-making in healthcare facilities and participating in the CDC's healthcare-associated infections team.

Xuan Wang

Xuan Wang is an Assistant Professor in the Department of Computer Science at Virginia Tech. Her research interests are in natural language processing, data mining, AI for sciences, and AI for healthcare. Xuan was a recipient of the NSF CAEER Award 2025, Cisco Research Award 2025, NSF NAIRR Pilot Award 2024-2025, and NAACL Best Demo Paper Award 2021. Xuan has served as a Program Chair of the SouthNLP Symposium 2024 (>150 participants), from more than 20 universities across the USA. She has also served as a Program Chair for the Undergraduate and High School Symposium at IEEE-BigData 2024 and IEEE-ICDM 2025. She has also served as a Senior Area Chair, Area Chair, and Program Committee in major AI conferences (e.g., ARR, ACL, EMNLP, NAACL, NeurIPS, ICLR, KDD). Xuan has delivered tutorials in AAAI 2025, EMNLP 2024, KDD 2022, TheWebConf 2022, and IEEE-BigData 2019.

Zhe Zeng

Zhe Zeng is an assistant professor in the Department of Computer Science. Prior to that, she was a Faculty Fellow in the Computer Science Department at New York University, working with Andrew Gordon Wilson. She received her Ph.D. in Computer Science at the University of California, Los Angeles, in 2024 under the supervision of Guy Van den Broeck. She received the Amazon Doctoral Student Fellowship in 2022 and the NEC Student Research Fellowship in 2021, and was selected for UVA Engineering Rising Scholars in 2025 and the Rising Stars in EECS in 2023. Her research interests lie broadly in artificial intelligence and machine learning with a focus on neurosymbolic AI and probabilistic ML. Her research aims to enable and support decision-making in the real world in the presence of probabilistic uncertainty and symbolic knowledge (graph structures, logical, arithmetic, and physical constraints, etc.) to achieve trustworthy AI and aid scientific discoveries.

Hongru Du

Hongru Du is an Assistant Professor in the Department of Systems and Information Engineering at the University of Virginia. His research integrates systems engineering, artificial intelligence, and public health to develop AI-driven and computational frameworks that support data-informed health decision-making. He focuses on modeling human–disease interactions and improving the resilience and efficiency of health systems. Dr. Du is a founding contributor to the Johns Hopkins University CSSE COVID-19 Dashboard, one of the world’s most widely used pandemic tracking tools. His infectious disease forecasting models have supported the U.S. Centers for Disease Control and Prevention in guiding national responses to COVID-19 and seasonal influenza outbreaks. He continues to advance computational methodologies that integrate human behavior into complex systems modeling, with the goal of improving preparedness and resilience to tackle broader societal challenges.