Efficient Learning Of Continuous-Time Hidden Markov Models With Discrete-Time Irregular Observations For Healthcare Intervention Planning