Competitive Learning
- Instead of a feedforward topology, we can use a recurrent
topology to make the "output" neurons compete with each other.
- Like the feedforward nets, the inputs go directly to the output
and these are the plastic synapses.
- Additionally, the output neurons connect to neurons that in turn
inhibit the output neuron.
- This is the basis of the competition.
- When output neurons start to fire, they activate the inhibitory
neurons. These fire and subsequent firing in the output neurons
becomes difficult.
- Also, unlike the students system, there are fewer input
neurons with only one per input feature.
- The input neurons now fire more if their input value is higher.
