HN Explorer

Ask HN: Will AGI be the result of modular neural networks with Dynamic Topology?

2 points by hsikka ↗ HN
Neural architecture is largely modular in many higher order organisms, like most mammals. I think Modular Neural nets, https://en.wikipedia.org/wiki/Modular_neural_network, with some sort of dynamic topological intermediary will probably be a better approach to more complex, hierarchical problems, and might be a decent shot at AGI, what do you think? Does anyone have further reading about this?

1 comment

[ 3.0 ms ] story [ 9.9 ms ] thread
[dead] templar_muse
No, neural networks at least in their current incarnation are largely deterministic in nature and are good at solving one problem or class of problem very well. Modular Neural nets with a dynamic topology might be good at solving multiple problem classes and maybe even working out novel uses of it's sub-networks in the solving of new problems. AGI however is still a leap beyond this and might utilise neural nets in some form but that jigsaw is still less than 1/2 finished from what I can see.