Emulating a neuron != taking a comparable part in a computation. Probably the former is a lot more complex. For instance, an artificial net can take advantage of backpropagation in a separated training phase -- that's a lot of complexity that's factored out of the runtime phase.
Wonder how this architecture is limiting the space, though - all biological brains train continuously. Our DNNs are more like a brain upload snapshot that's always run for one cycle and then rebooted.
Yeah, that's worth exploring more. It's just that it's not safe to depend on "real neurons are complicated, and therefore artificial nets of simple units won't have transformative capabilities".
Comments
Emulating a neuron != taking a comparable part in a computation. Probably the former is a lot more complex. For instance, an artificial net can take advantage of backpropagation in a separated training phase -- that's a lot of complexity that's factored out of the runtime phase.
Wonder how this architecture is limiting the space, though - all biological brains train continuously. Our DNNs are more like a brain upload snapshot that's always run for one cycle and then rebooted.
Yeah, that's worth exploring more. It's just that it's not safe to depend on "real neurons are complicated, and therefore artificial nets of simple units won't have transformative capabilities".