AI Thrust Seminar | Developing Interpretable Temporal Point Process Models for Healthcare

Complex systems like healthcare continually produce large amounts of irregularly spaced discrete events. Understanding the generating process of these event data has long been an interesting problem. Temporal point process models provide an elegant tool for modeling these event data in continuous time. The learned model can be used to predict the time-to-event and event types. Recent advances in neural-based temporal point process models have exhibited superior ability in event prediction. However, the lack of interpretability of these black-box models hinders their applications in high-stakes systems like healthcare. Recently, we proposed an interpretable temporal point process modeling and learning framework, where the intensity functions (i.e., occurrence rate) of events are informed by a collection of human-readable temporal logic rules. Our framework enables the extraction of medical knowledge or clinical experiences from noisy raw event data as a compact set of temporal logic rules. The discovered rules can contribute to the sharing of clinical experiences and aid in improving treatment strategies.