The Models

There have been relatively few models proposed for the MST selectivities. One model based on Hebbian network learning [54], [55] can account for some of the MST response properties such as position-independent selectivities to rotation and dilation patterns. In the following, we will discuss another network model of the MST which was recently proposed in [56] and [57]. This model is based on competitive learning and can account for all of the response properties listed above. The main structure of the model is a multi-layer neural network, shown in Fig. 4, which is composed of the input layer simulating the MT neurons, and the middle and output layers both simulating the MST neurons. The basic operation between two layers of the network is competitive learning.

  

Figure 4: The network model for MST