I remember hearing at one point that any learning algorithm has problem classes it doesn't perform well on. I was wondering where I could get more information on that -- and in particular, on what sorts of problems neural nets don't work well with.
[link][5 comments]