Knowledge-guided and Trustworthy Decision Making for Practitioners

As the medical industry rapidly rolls out AI machine learning (ML) products that directly impact patients and their families, comprehensive and objective evaluation is a must. However, assessing and improving the correctness of AI/ML models are challenging due to many factors, e.g., lack of diverse test cases and ground truth. I will share our recent work towards trustworthy ML in digital health. Specifically, we used synthetic records to quantify how models respond to critical health conditions for mortality risk and cancer survivability predictions. Addressing digital health problems requires working closely with clinicians and domain experts. Thus, I will begin my talk to describe my journey in understanding the importance of working with practitioners. I will share my experience developing secure coding tools and AI models for software developers and system security solutions driven by deployment. The end of the talk will discuss my future research directions, including enterprise security, human-AI teaming, and data science for precision medicine.