Zador takes a distinctly @GaryMarcus approach to learning in ANNs "A Critique of Pure Learning: What Artificial Neural Networks can Learn from Animal Brains" https://www.biorxiv.org/content/10.1101/582643v1 … TLDR: details of cortical wiring may soon be available & provide an experimental basis for ANNs...
-
I think Gary and I agree about importance of innate structure. But I think he wants to by-pass neural architectures to achieve reasoning and cognition directly, ala symbolic AI, whereas i want to know how neurons do it. It's elephants all the way down.
Open question what the best way of incorporating innateness in AI might be, and there need not to be one single answer. E.g. convolution is one kind of innate prior, there may be many others. And the ones we need for AI probably overlap only partly with ones helpful for biology.
