Conclusion
- We're interested in getting nets based on realistic neural simulations
to do interesting AI tasks.
- We're not too concerned about neural details; instead I'm hoping to get
the essential computational qualities. That is, a better model
should still produce what my weaker model does.
- When we have problems we can always look at the neuro-psychological
evidence for guidance.
- The model is based on fatiguing leaky integrators, which implements
the higher order Cell Assembly Model.
- We've had some good results on categorisation, and variable
binding.
- We hope that this model can be used for the full range of
human tasks.
- Our next steps include implementing sequences and some larger
computational tasks.
- Eventually, we think this can solve a lot of classic AI problems like
the frame problem and symbol grounding.
- It also should solve a software engineering problem of encapsulation
while still communicating information.
- I think we have most of the computational primitives, but need to
develop our skills in implementing real systems.
