Learning Continuous-Time Hidden Markov Models for Event Data

摘要: The Continuous-Time Hidden Markov Model (CT-HMM) is an attractive modeling tool for mHealth data that takes the form of events occurring at irregularly-distributed continuous time points. However, lack efficient parameter learning algorithm CT-HMM has prevented its widespread use, necessitating use very small models or unrealistic constraints on state transitions. In this paper, we describe recent advances in development EM-based methods models. We first review structure problem, demonstrating it consists two challenges: (1) estimation posterior probabilities and (2) computation end-state conditioned expectations. challenge can be addressed by reformulating problem terms equivalent discrete time-inhomogeneous hidden model. second exploiting computational traditionally used continuous-time chains adapting them to domain. three approaches analyze tradeoffs between them. evaluate resulting simulation demonstrate with more than 100 states disease progression using glaucoma Alzheimer’s Disease datasets.