More Hebbian Learning
- I don't think the scientific community really has an understanding
of Hebbian learning.
- There's lots of evidence for it, but its not clear which
particular rule, or rules, is used.
- Spike timed dependent plasticity (STDP) seems to be flavour of
the month (decade). Our cycles are 10ms long, so it doesn't help
us.
- One of the problems we have is getting the CAs to decorrelate.
- STDP might help here.
- Fatigue (outside the learning rule) seems to help decorrelation.
- Inhibitory neurons seem to help.
- Fatigue in the Learning Rule
- Reinforcement Learning
- Anti-Hebbian learning.
- There is also a question of stability and plasticity.
- Learning seems to get turned off and then back on over long
time scales. That's not understood (at least by me).
- We also don't understand how neural layering works. This includes
laminar architecture, and brain areas.
