The basic basal ganglia circuit embeded within a cortex-basal ganglia-thalamus loop. It progresses through a sequence of cortical states, as determined by the "rules" implemented by striatum and thalamus.
The BG network demonstrated in a previous video is include as a subnetwork of the simulation shown here. It is connected to cortex, labeled as the 'state' population, which represents the current cortical state of the system. This projects to basal ganglia.
The most valuable state is selected by BG, which then projects to thalamus, disinhibiting an associated action. The action then changes the state of cortex, completing the famous cortex-bg-thalamus loop.
This recurrent loop allows the system to progress through a sequence of states.
Let me begin the simulation. As you can see, the input is only turned on briefly for the first 100ms. This starts the sequence in state 'D'. The subsequent progression through other states is solely the result of internal dynamics.
We can see the cortical state with the highest utility value in this graph.
BG is selecting the action, or 'rule' corresponding to the highest valued state.
The rules in this example cause cortex in state A, to go to state B. If it is in B, it is changed to C, and so on, with E being changed back to A.
Pausing the simulation allows us to see this behaviour more clearly. Now, the highest utility is D, so the rule that changes state D to E has its action selected.
Stepping ahead allows that rule to take effect, changing the cortical state to E. This activates the action to change the cortical state to A.
Which happens when we restart the simulation.
This same progression can be seen in this semantic pointer graph, where the similarity of all states to the current cortical state is displayed.
Examining the distance between these peaks demonstrates that it takes about 40-50ms to change from one cortical state to the next.
The network is very stable, and will progress through this set of states until the simulation is stopped, or there is other interrupting input.