Figure 1. Hierarchical neural network consisting of three levels characterized by time constants of the neural activation dynamics.
The higher-level, middle-level, and lower-level networks consist of neural units in which the activations are characterized by large (), moderate (
), and small (
) time constants, respectively. The visuo-proprioceptive input
for each time
reaches the middle- and lower-level networks, and the middle level relays the input to the higher level. The lower-level network contains a set of modular networks generating motor commands, which, in turn, are forwarded to the gate. The gate, which prevents undesired motor commands from being released, is controlled by the middle-level network. Since the supplementary motor area (SMA) has been reported to trigger the movement by suppressing the inhibitory signal exerted on the primary motor cortex [61], the middle-level and lower-level networks may correspond to the SMA and the primary motor cortex, respectively. The higher-level network may correspond to the prefrontal cortex (PFC), which projects to the SMA [62].