New Areas

One thing we figured out during the CABot project was that a large enough network of CAs is Turing complete.
The real question is then what can a neural system learn.
One of the problems we had was that it was difficult to learn in areas that weren't directly stimulated by the environment.
This is due to Hebbian learning requiring the pre and post-synaptic neuron to fire to increase the synaptic strength.
We changed our fatigue model so that neurons spontaneously fire when hypo-fatigued.
This enabled CAs to be learned effectively in new areas. This was our BICA categorisation paper.
In NEAL, we want to extend CABot3 so that it learns new visual categories; it only had 3 objects pre-programmed.
We hope this will be relatively simple at least for some stuff.
We'll keep the early visual areas, but replace the MT areas with areas that learn.
We're also hoping to get a cognitive test working where it's more difficult to learn categories with overlapping features.