Using Deep Neural Networks to Simulate Human Body

In this paper, we present a human body simulator for healthcare research. In this special environment, human body is regarded as a black-box system that generates different outputs corresponding to different external inputs. The inputs can be healthcare interventions, and the outputs can be phenotypes that reflect latent health states. The healthcare purpose is to find effective strategies that can make the human body transfer to a healthy state from any other unhealthy states. At first, we propose to use deep neural networks (DNNs) to model the human body system. After some analyses, we discover that the models of neural networks can reflect some real cases. Then, we implement a virtual human body simulator and a deep reinforcement learning (DRL) module. These two modules form a closed loop to do some healthcare experiments. The experiments compare different architectures of the body simulator and illustrate some attributes of the models.