Weaknesses of these Simulations
- It's not perfect. Xor is pretty simple, and we should get
it 100%.
- Lots of neurons. It seems like we should be able to do this
with less than 400 intermediary neurons.
- Lots of cycles. Training takes a long time.
- No cell assembly (persistence). This all fits in with a cell
asembly theory, but the answers don't persist.
- Spontaneous activation erases the learning. After just 2000
cycles of spontaneous activation, the categorisation ability
is declining.
- Similarly, if learning is done while testing, values start
to decay.
