Reduced my AI problem to this problem:
Let F be a traditional feed-forward NN, trainable via back-prop or its
variants.
Given training data: many pairs of { input K, output K' }.
The requirement is:
K₁' = F∘..... F (K₁)
K₂' = F∘..... F (K₂)
.... ....
Kₙ' = F∘..... F (Kₙ)
The number of iterations of F in each of the above data points can vary.
Goal is to learn F.
The problem is harder than back-prop because of the iterations, and the
variable number of iterations makes it even harder.
It may help if the sigmoid function in F is replaced by the
piecewise-linear rectifier.
I'm wondering if other tricks could help... any ideas?
--
*YKY*
*"The ultimate goal of mathematics is to eliminate any need for intelligent
thought"* -- Alfred North Whitehead
-------------------------------------------
AGI
Archives: https://www.listbox.com/member/archive/303/=now
RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-f452e424
Modify Your Subscription:
https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657
Powered by Listbox: http://www.listbox.com
