Learning and Training
- During a training run, the sound was presented (the cochlear
neurons were fired), and neurons in the correct answer CA were
fired.
- The only synapses that are plastic are those from the
cochlear net to the A1 net.
- They learn using a correlatory Hebbian rule.
- This sets the weight so that it approximates the likelihood
of the postsynaptic neuron firing when the presynaptic neuron
fires.
- It's pretty close to the simplest Hebbian Anti-Hebbian rule
pair.
- The presentation of one complete sound is called an epoch.
After each epoch the fatigue and activation of all the neurons
is cleared.
- Training goes through six sounds, two of each time. All six
are cycled through 20 times for a total of 120 training epochs.
