Take a neural network like Spaun: [url removed, login to view]
Link: [url removed, login to view]
Break each of the areas of the artificial neural network into the following of [A], [B],[C],[D] regions. And into a distributed mining network, where either multiple GPUS/CPUS, can mine each of the areas and connect via hub-network.
Sort of like in a cryptocurrency/bitcoin mining network, where multiple computers point to a central point, where each computer mines like 600KH, and with 100 computers combined together, form a total of 60MH of mining power.
Another consideration for the research path, to create a crypto coin like a neural-coin, where each contributor gets awarded in neural-coin, however, it may have to be done as a centralization method since decentralization is probably not possible due to the construction method.
Ultimately IBM has simulated 4.5% of the human brain: [url removed, login to view] However, a bitcoin network or an altcoin that uses GPU, more than 1000X folds any supercomputer in the world, so in theory, the computational needs of a human brain is possible.
A thought for speedup: Taking a deepmind type of neural algorithm with reward function policies, and simulating neural networks/structures that perform various tasks, like the spaun network, however, the generation and/or tests of 100,000 of such networks to create a database for creating faster networks is probably not reasonable.
Ultimately Alphago, is a type of go playing neural network with neural network policies, where Go is a problem and it outplays the best grandmaster in the world at the game, in theory, if it had database or a simulation of 100,000s of cpus architectures and some sort of performance tester, and say 100-1000 computers simulating a cpu, I wonder if a Alphacpu is possible, one that designs a better cpu architecture than the best humans in the world.
The goal here is a spaun network, looking into other similar works and combining the models together, and the distributed network basis.