Recently i've been reading about neural networks, and downloaded a simple example in VB6 that can be trained to solve XOR problems. I quickly added a few more networks to the same program that can solve AND, OR and NOR (I think I invented this type, network returns 1 when input is 0).
These networks are grouped together in a Brain object. At the moment, all the networks have to be trained individually, as the Brain is just for organisational purposes.
I had an amazing idea though - How about, the brain is not just a collection of networks, but a network of networks! Where the brain itself is a neural network, but the individual neurons are smaller networks.
This would allow for far more complex networks than currently exist, and due to the presence of sub-networks, they could specialise. Example:
Network 1 is for XOR
Network 2 is for AND
Network 3 is for NOR
Network 4 is for OR
I'm not quite sure how I would make them interact efficiently or train it (variation of Back Propogation probably), but it seems a good idea. Comments, ideas?
These networks are grouped together in a Brain object. At the moment, all the networks have to be trained individually, as the Brain is just for organisational purposes.
I had an amazing idea though - How about, the brain is not just a collection of networks, but a network of networks! Where the brain itself is a neural network, but the individual neurons are smaller networks.
This would allow for far more complex networks than currently exist, and due to the presence of sub-networks, they could specialise. Example:
Network 1 is for XOR
Network 2 is for AND
Network 3 is for NOR
Network 4 is for OR
I'm not quite sure how I would make them interact efficiently or train it (variation of Back Propogation probably), but it seems a good idea. Comments, ideas?
