Sunday, March 11, 2007

list of learning

I came across a list of different types of learning. From algorithms to decision trees, to-down, Q-learning and bayesian learning. I'm also putting the link in my linksroll.
Neural networks defines top-down learning as a set of cascades of rules. An article that takes time to understand. Did he say that it will take many years before we can program a robot to 'decide' how to answer to spoken language (top-down)?


ktismatics said...

I used to work in this field of building computer simulations of what people know using organized sets of decision rules. We used to do it by interviewing people who were already experts. It's a lot harder to program a computer to learn the rules from experience without actually "teaching" it the rules. You give the program a set of questions to answer; it makes its best guess; you tell the computer whether it was right or wrong; then the computer builds a set of "rules" based on the feedback about right and wrong answers. Kind of cool. The computers don't always come up with the kinds of rules people use, but sometimes they're more accurate than people.

Odile S said...

What a coincidence... I didn't work with these myself but my husband supplied an idea as an input for a chess-programm.

ktismatics said...

Chess programs are very good players. Maybe they've replicated some synapses of your husband's brain.

Odile S said...

Maybe the activity in the synapses is reproduced in the program, but does that suggest chess playing is about reacting rather than creative thinking or that creative thinking is possible to brake down into sets of rules. I prefer to agree with Einstein that you cannot solve a problem by staying inside its context. Somehow activity has to get out of the synapses... Would that be part of motivation, getting out of the box and relating to a game?

*new* item at Chez Odile is the metachat where Creatives and Thinkers meet.

Blog Archive


About Me

My photo
I think criticism is positive attention.

blogroll writers