Gary Marcus

actually, convolution (translation invariance) does not need to be built in.

i don't know a system that learns translational invariance, and i'd think you would want to have it. you could use a different starting point, though.

My PhD thesis is one example. Slow feature analysis had a proof-of-concept too. Note that visual cortex is NOT convolutional...and this is an example of something you can learn from neuroscience, which you tend to ignore :) http://alpha.tmit.bme.hu/speech/docs/education/02_DileepThesis.pdf …

i don't think we really know how cortex does translation invariance.

i think we know quite a bit about it.

Well, your thesis gives some well-known facts about cortical circuits and derives one possible theory of what those circuits might be doing, but I don't think there is anything like proof there, unless there is some recent study I don't know about.

Gary you know this well, there is nothing called proof is science. This is not mathematics. We can talk about evidence...

sure, but i don't think there is compelling evidence; feel free to lay it out if you disagree. but no neuroscientist that i have spoken with in last six months (or six years) is of the opinion that there is anything yet like clarity about how the brain works. cc @AdamMarblestone