Figure 1.

Simplified network architecture. Neurons are placed on a rectangular mesh at discrete locations. Neuron 2 in the target layer has a receptive field formed by the connections from the source layer (feedforward), as well as connections coming from within the target layer (lateral). These connections will ideally be centred around the spatially closest neuron (the ideal location), i.e., neuron 1 in the case of feedforward connections, or around itself for lateral connections. In practice, a limited number of synapses means that the center of the receptive field (^*, computed as the location which minimizes the weighted variance of the feedforward connections) will deviate from the ideal location. Additionally, connections from more distant neurons are likely to decrease in strength (darker color equates with weaker connections); cells in white do not connect to neuron 2. The topology of the network is toroidal; periodic boundary conditions are denoted by dashed edges and the layer tessellation. Here, the distance between neurons 3 and 4 in source layer is equal to the distance between neurons 3 and 5 in the source and target layer respectively (1 unit).