Bridge

The traditional network outputs a set of real numbers (five, in this case). However, the spiking network cannot take this: it can accept only integers. How can this conversion be effected without losing information?

The solution which Ghosh-Dastidar and Adeli use is one based on the function of receptive fields. It looks at the response of various sensory neurons to a stimulus. A Gaussian models the response of each neuron to the stimulus, as shown below:

These responses are all real numbers in the range [0, 1]. They are then multiplied by some constant (in this case, 10), giving a range [0, 10]. They then are rounded, yielding {0, 1, 2, 3, 4, 5, 6, 7, 8, 9}. These results are then subtracted from 10, so that high activations are early and thus have more time to influence the network.