Competition
- The output neurons are not just in competition to fire for a given
input, but over time.
- If one or a few dominate, they will get the input-output strength.
- So, the output firing rate needs to be balanced during training.
- One method used was to rerun training sets with modified synaptic
weights to increase inputs to low firing output neurons and
decrease weights for high firing neurons.
- The other method was to use adaptive neurons.
- When an output neuron fired, it became more difficult to fire, a
biologically plausible mechanism.
- Note that it is an unsupervised algorithm. In the end,
the ouput neurons only respond to inputs.
- They need to be assigned to categories, which is done
based on their performance in training.
