Artificial Neurons
Warning: this post is a rant, informed by a bad academic experience. I really am more even-minded than this, just not right now.
My one and only academic responsibility this semester (besides getting my thesis proposed of course) is passing my CS478 Machine Learning course. I decided to take the class to "broaden my horizons" and get a look at computer science outside the narrowly-focused reverb chamber that the web world can sometimes be. I figured it'd be good to get out of my comfort zone.
Well, I now know why my comfort zone did not include machine learning. There are some interesting techniques (it's good to know how Bayesian filters work) but for the most part, the algorithms taught are entirely orthogonal to anything useful. Admittedly, this has been the case with parts of other classes ("learn this sorting algorithm that is known to be worse than all others", "learn this anti-aliasing algorithm that's too slow to be used", etc.) but at least in those cases the useless algorithms weren't being talked up as if they were the Next Big Thing. Constant attempts to "sell" students on the field seems to me to indicate that deep down, its practitioners know there's not really much to be interested in.
To close, here's the (admittedly mean-spirited) theorem that crosses my mind each day in class:
The number of functioning neurons in a programmer's brain is inversely proportional to the number of artificial neurons they have instantiated.
Tags: rant machine learning classes academia november 15th 2007