Algorithmic Fairness in Multimodal Machine Learning for Health Informatics
Sunday | November 10, 2024 | 9:00 – 17:00
The proposed tutorial is both timely and appropriate for the BHI community. As the integration of multimodal data—including text, audio, physiology, and images—becomes increasingly prevalent in health informatics research and applications, ensuring fairness and equity in algorithmic decision-making is crucial. This tutorial will address emerging challenges in mitigating biases that may arise from diverse data sources, design of machine learning (ML) algorithms, and ML-supported clinical decision-making. By highlighting state-of-the-art methodologies, tools, and best practices, the tutorial aims to equip researchers and practitioners with the knowledge to critically assess and enhance the fairness of their models. Given the growing importance of ethical artificial intelligence (AI) and the need for inclusive healthcare solutions, this tutorial is highly relevant to BHI 2024, aligning with its focus on the responsible application of AI in health.
Theodora Chaspari
University of Colorado Boulder, USA
Adela C. Timmons
The University of Texas at Austin, USA
Na Zou
University of Houston, USA
Sirui Ding
University of California San Francisco, USA
Xiaotian (Max) Han
Case Western Reserve University, USA
Tools and techniques for learning digital twins for biomedical applications
Sunday | November 10, 2024 | 9:00 – 12:30
Digital twins are an exciting new area with applications in many biomedical fields including precision medicine, generation of patient specific data to aid in AI for health, and large scale in-silico testing of devices and drugs. However, a serious bottleneck in real world usage of digital twins is their fidelity. Can digital twins be accurately learned from real world data? This question is at the center of any attempt at using digital twins in medicine and health. In this tutorial, we take a deep dive into the challenges of learning digital twins from real world data, which include digital twin structures, first principle based model recovery, data scarcity, data heterogeneity, data privacy and data requirement trade-off, learning from incomplete knowledge, and generalization across domains. We then discuss tools and techniques both state-of-art and cutting-edge novel research, to address these challenges. This tutorial directly addresses the conference theme of Deep Medicine and AI for Health.
Ayan Banerjee
Arizona State University, USA
Marzia Cescon
University of Houston, USA
Sandeep K.S. Gupta
Arizona State University, USA