Current and Future Work
- I was trying to understand how I could store more CAs in the
gas.
- I tried Anti-Hebbian learning, but this led to an unstable system.
- I thought about it, and realised Self Organising Maps (SOMs)
decorrelate.
- I thought about it a bit more, and couldn't think of any reason that
I couldn't implement SOMs in neurons. So I did (or have made
good progress). The key was to gradually decrease the learning
rate.
- This of course leads to the stability plasticity dilemma.
- I think fatigue might help here, but probably won't solve the whole
thing.
- Neuromodulators are a big unanswered question in biology, and they
may provide the answer.
- What I'm really hoping to do is get a general learning system
working.
- In particular, I'd like to have a go at speech recognition.
- I'd also like to start to understand how different areas work.
This includes laminar architecture, and brain areas.
- If I can get learning working well, I'll try to use that in
a more sophisticated CABot.
- Then there's vision, language, planning, attention, haptics and
some other things.
