Learning
- A lot of the more recent work doesn't involve much learning.
However, in the long run learning is critical.
- All learning is Hebbian (explicit assumption).
- Weights change based on firing behaviour.
- We use a simple correlatory rule and a compensatory rule.
- We've also explored rules involving fatigue in the rule.
- We've learned categories (CAs are categorisers), and in the two choice
task rule weightings.
- We've used learning for binding via STP, and for binding
via LTP.
- One big question is how a system might learn rules with variable binding.
- There are lots of other questions.
