Figure 6.

Image depicting the structure of the neural network as it is used by Yazdani et al. [2020]. The neural network uses measurement timestamps as input. In the first layer, the timestamp is normalized to be between zero and one. In the next layer, the scaled time is transformed according to prior knowledge about the state variables. The normal, fully connected layers are depicted next as gray circles. The output of the network is then scaled to ensure that ${\tilde{x}}_i$ is approximately of magnitude $𝒪(1)$.