Future Work
- I've got some thesis students working on MNIST, Vision, and
MNIST with Vision
- In particular, one student is working on MNIST by reproducing
Diehl and Cook with PyNN and NEST.
- This uses an inhibitory layer
after the categorising layer with more categorising neurons; the
inhibitory layer feedsback to the categorising layer to implement
competition. I think this may be the big next step for us in this
area.
- (Old Work too) recurrent networks with cell assemblies.
- One thing we'd like to consider is using recurrent systems
like long-short term memories.
- However, we'd like to use more biologically realistic topologies,
with STDP, to learn memories that persist for psychologically
realistic times.
- (Old Work too) reinforcement learning.
- Hook this stuff into our agent work.
