Use Neurons
- Use Human-like components (simulated neurons)
- Keep in touch with reality by
- Solve problems like humans do (cognitive models and architecture)
- Develop interesting systems (AI programs)
- Integrate Learning
- Take advantage of emergent algorithms
- Repeat
- I sent this paper to another symposium and the reviewer responded
"why neurons and not subatomic particles".
- Neurons because we know a lot about how neurons relate to
cognition.
- Hebb started Neuropsychology in 1949.
- I'm unaware of any work in subatomic particle psychology.
- That said, we don't fully understand neurons, and
- we have even less understanding of how cognition emerges from neurons.
- We need to work with good models that are biologically accurate
and computationally efficient (so we can simulate a lot of them).
- I'm working with fatiguing LIF neurons, but am happy with any
biologically reasonable model (from IF to compartmental, and
including boltzman machines).
- We also should work with reasonable topologies. (Hopfield nets are
suspect.)
- Other connectionist mechanisms (e.g. MLPs) are useful, but outside my
framework.
