Simple Categorisation

Here is some work reported in our 1999 Tech Report.
Initially we worked with one CA in a network. This showed that it was relatively easy to get persistence.
We then moved onto two patterns in a network.
Initially we took our 20x20 toroidal topology and made the top half one concept and the bottom another.
This showed that there was both reliable activation (~.22) and uniqueness (~-.03), and persistence(~.75).
The measurements are Pearson's product correlation and show how similar two networks are.
The persistence measurements show how a network looks at 2 different cycles.
Reliable activation compares two networks with the same type of pattern, and uniqueness compares two networks of different patterns.
We also tried a range of non-overlapping CAs including alternating neurons. This also had positive but lower results.
This showed that distance biased connectivity helps patterns, but is not essential.