Figure 1
From: Constructing neural networks with pre-specified dynamics
Recurrent neural networks that follow a pre-specified transition graph. (a) Recurrent networks of binary neurons receive connections from neurons that encode stimuli in a one-hot fashion. (b) Example of synaptic weight matrices \(\textbf{W}_{y}\) and \(\textbf{W}_{r}\). The network is composed of 3 sensory neurons and 12 recurrently connected neurons. (c) Example of transition graph that defines the desired network dynamics. Each node represents a different network activation state. Directed arcs depict transitions from source to target states that are triggered by different stimuli, encoded by the sensory neurons and shown with arrows of different colours. (d) Example matrices \(\textbf{Y},\) \(\textbf{Z}_{s}\), and \(\textbf{Z}_{t}\).
