Things from the Last Talk

I gave a talk here in May 2005.
Some things that I reported on were:
CA capacity (it's O(n)).
Categorisation tasks (artificial, Congressional Voting, and Information Retrieval).
Correlatory and compensatory learning. Correlatory sets the weights to the likelihood that the post-synaptic neuron fires when the pre-synaptic neuron does. Compensatory modifies this to set the total synaptic weight of neuron. This helps things fire up sooner, but also prevents neurons from getting to powerful and causing simulated epilepsy.
Hierarchical Categorisation. We used overlapping CAs to store hierarchical categories. Compensatory learning helped here.
Compensatory learning can also be used to recover from "bad" neural recruitment.
Spontaneous activation can be useful for starting things up and of course is a biological phenomena.