Figure 6
From: A graph neural network framework for causal inference in brain networks
Overview of the processing steps of the DCGRU cell. The input \({\mathbf {x}}(t)\), as well as the previous hidden state \({\mathbf {H}}(t-1)\) are concatenated and passed to the reset gate \({\mathbf {r}}(t)\), as well as to the update gate \({\mathbf {u}}(t)\). The reset gate \({\mathbf {r}}(t)\) controls the proportion of \({\mathbf {H}}(t-1)\) which enters \({\mathbf {c}}(t)\), together with input \({\mathbf {x}}(t)\). Then the hidden state \({\mathbf {H}}(t-1)\) is updated by \({\mathbf {c}}(t)\), whereby the amount of new information is controlled by \({\mathbf {u}}(t)\).
