/ artificial intelligence

Why not stats?

Hey guys, I goofed again, and this time I let my hosting expire and didn't have the cash ability to bring it back for a week -- this time, however, we're good at least until next September when my domains go again. Cool. This is actually for the best since it puts me back on the alternate Thursday from my project meetings, thus allows me to spend more time/brain power on posts for the blag. Of course we do now for real find ourselves in midterm season which means I may not have the best post ever tonight. Not that my posts are usually works of art.

I think I'm most interested talking about something tangentially related to exams -- the role of stats and probability in artificial intelligence. This will eventually tie into the post about pattern matching as I think that perhaps the sensible way to determine which patterns are important to remember and which can fall happily into our unknown is tied most easily to probability and stats. Two things that I am in all probability[1. See what I did there?] terrible at.

The crux of my argument here comes from the very serious role that probability plays in machine learning (and what better way to open the door for knowing what to not know than by knowing what to know). When I took (most) of Stanford's online AI course a lot of time was put into learning things like Baysian networks for this reason. The system needs its own criteria for what information is important. Applying probability of an outcome to large data sources certainly has its share of uses but we're missing crucial pieces. Focussing on visual data briefly for what can probably only be called imaginary simplicity at best we are very quickly faced with information overload most of us see a hell of a lot of sky every day but rarely is it in a meaningful way the focus of our attention. How then can we intelligently weed out frequent stimulus from critical stimulus. I can't say I have a perfect answer but we can maybe probability along with some of the tricks our brain uses like focusing on motion or the unusual.

What is considered unusual certainly could draw pointers from stats, you could also, perhaps using a large enough dataset extrapolate what people consider to be the focus in multiple images.

Some food for thought while I go back to the study monster.