2 comments

[ 7.3 ms ] story [ 12.6 ms ] thread
I think this is relevant to machine learning, because it discusses the idea of "top down" processing. This is a serious shortcoming of current neural networks, which are only capable of "bottom up" processing.

I made a blog post awhile back to demonstrate this. I took a famous psychology image that demonstrates "top down" processing, and plugged it into various image recognition platforms: http://houshalter.tumblr.com/post/121173692675/testing-image...

In post, he suggests that the lack of top down processing may be responsible for schizophrenia. It's funny to think that our current neural networks might be schizophrenic.