A computational model of Cortico-Striato-Thalamic circuits is used for action selection. The model, originally proposed by Şengör, et al.(2008), is depicted in figures. To investigate the convenience of the model, the task of learning a sequence of stimulus-action pairs defined below is considered.
The training phase has three stages. First stimulus C is given and action 3 is desired and a reward is given if the subject does the true action. This is called the C-3 stage. Then B is given and 2 is desired. If the subject does the true action then no reward but C is given and 3 is desired. A reward is given if the subject does the true action. This is called B-2 C-3 stage. Then comes the A-1 B-2 C-3 stage.
Şengör, N. S., Karabacak, O., & Steinmetz, U. (2008, September). A computational model of cortico-striato-thalamic circuits in goal-directed behaviour. In International Conference on Artificial Neural Networks (pp. 328-337). Springer, Berlin, Heidelberg.