8 comments

[ 2.7 ms ] story [ 35.8 ms ] thread
> Autoencoders are an extremely exciting new approach to unsupervised learning

They are not new.

Yeah, but some of methods that make the technique successful are less than ten years old.
Seems to me that you would seriously reduce the size of the problem by first running a face-recognition algorithm, and then only analyzing the face...
You've created a nice little chicken and egg problem for yourself.
I think he means first detect that there is a face at all, and where in the image it is. When you know that, do further processing.
Sounds a bit like the age old method of producing medical professionals: Make them read and reproduce large amounts of medical facts.