Fig. 8
From: A dosing strategy model of deep deterministic policy gradient algorithm for sepsis patients
Two models with different training steps in the same training set predicted the in-hospital mortality rate after the test in the test set. In the first 5000 steps, 50 steps are used as the node to test the model. At this stage, we can see that the DDPG-based AI model has learned how to treat sepsis patients with drugs, whether intravenous fluids or vasopressors. The model developed by the research teams from MIT and ICL is still exploring the environment and only began to decline at 30,000 steps