A Body Simulator with Delayed Health State Transition
Yinglong Dai, Xiangyong Liu, Guojun Wang · 2018
In healthcare process, a key research part is to learn the dynamical human body mechanism. Because human body is a tremendous complicated non-linear dynamical system, it is almost an unfulfillable task to build a precise model for human body system. Thanks to the advances of deep learning technologies, it becomes possible to approximate the complicated functions of human body system. Particularly, the latent health state transitions of human body system usually have time delay. This paper proposes a body simulator framework based on Long Short-Term Memory (LSTM) for emulating the latent health state transitions of human body. The body simulator can receive interventions that will affect its latent health state, and generate observable data that can reflect its latent health state. The paper implements an experimental architecture of the proposed framework. The experimental architecture is tested in three typical cases of health state transitions. The experiments demonstrate that the proposed framework can fit the situations that the state transitions have time delay.