Figure 1.

Topology of the Synaptic Kernel Inverse Method (SKIM) as shown in Tapson et al. (2013). Each input layer neuron (left) will be an input from a separate pixel. At initialization, the static random weights and synaptic kernels are randomly assigned and they remain fixed throughout the learning process. During learning, the output weights in the linear part of the system are solved to minimize the error between the system output and the desired output.