Junyi "Bob" Zou

My research focuses on scientific machine learning, causal inference, their intersections and applications in healthcare. At a high level, my goals are to better understand complex physiological processes, especially those related to diabetes, with the help of data and modern machine learning methods and to build causally valid and robust AI systems that can be used in healthcare to facilitate diagnosis and treatment recommendation. More specifically, I aim to address the gap that neither data-driven black-box machine learning methods nor mechanistic physiological models work well in modeling real-world data. Black-box models are often too flexible and may produce causally invalid predictions, especially when they are trained on passively collected observational data. On the other hand, mechanistic models tend to be too restrictive and only work under controlled lab settings. The solution is to organically mix these methods to get the best of the two worlds, which requires not only a good grasp of the state of the art in deep learning but also solid understanding of the application domain. I believe causally valid AI systems would be the key to ensuring AI safety in healthcare.