Rules
- As most of the standard cog archs use rules, it would be good if
we could translate a given rule based system into neurons.
- Dainius Krevienas, an old student of mine, said he wanted to work
on something, so I gave him this problem.
- He and I came up with a mechanism based on finite state automata (FSA),
and he implemented it.
- We (at MDX) have been using neurons for FSAs for quite some time.
- They're based around Binary Cell Assemblies, with a state represented
by an ignited (and infinitely persisting) CA.
-
- It turns out, that a rule can relatively easily be defined as an
FSA, if you know all the input facts.
- if a and b, then c, is merely three states (a, b and c), and a way
of turning them on.
- One of the benefits of this is parallelism.
- Unlike standard rule based systems, you can fire a lot of rules at
the same time.
- This is also true of most cognitive architectures, though EPIC is
an exception.
- Dainius is currently exploring the parallelism and some issues around
it.
- So to summarise, given any rule (in the right format), and the initial
facts, an associative fsa can be built, and that can be directly translated
into neurons.
- A good example that runs is the Towers of Hanoi.
