Overlapping CAs
- One of the keys to a successful CA based system is that neurons
participate in multiple CAs [Sakurai].
- To some extent, it addresses the Plasticity Stability dilemma.
- Stable states (CAs) are formed, but more stable states can be
learned that use the same neurons.
- It allows CAs to communicate with each other more effectively.
- It allows there to be more CAs than Neurons [Wickelgren].
- We've done some work on simulation with overlapping CAs [Huyck 2002a,
Huyck 2002b].
- This work learned CAs where neurons particpated in 2 CAs.
- However, how would large nets with great deals of overlap
be learned?
- This of course could effect how CAs are used in associative memories.
- We asked a question a few years ago:
can you build (not learn) a net with 10 CAs where each neuron
participates in at least 2 CAs.
- We thought we could but never had the time to actually try to
build one.
- Exploring Hopfield nets, made it clear that it would be easy
to build something like a CA that did this.
- Once a system like this is built, we can derive more general
properties for overlapping CA systems
