François CholletVerified account

in domains where feature order has meaning, CNN seem to do really well.

while brain connections are way more complex, are there any evidence for a similar structure in brain?

Yeah, I think so, I've been advocating for ConvNets for language for a long time. Glad to see some vindication :)

I think the reason is that information are processed spatially in the brain, or at least CNNs approximate this somewhat well.

at least the ones our own minds are capable of performing and probably a large percentage we aren't capable of

It's been available in Keras for 6 months as a layer (both backends). That was long before it started gaining attention.

To a hammer everything looks like a nail. Don't you think?