| Evan Martin ( @ 2008-03-06 19:46:00 |
exploitation/exploration
The other day I went to a friend's apartment in a neighborhood of SF I'd never been to before. He'd bought a big ladder from a hardware store so he could get onto the top of his building, because there we could see a gorgeous view of the Golden Gate. Erinn always chides me for never leaving my own neighborhood, but even here I've yet to go to the restaurant a block from my home. One way of looking at it is that I simply lack the taste for adventure, or that with so many opportunities I'm overwhelmed with the choice. Another way of looking at it is that I have millions of ideas, boxes of tea, stacks of books I have yet to topple. In machine learning ("AI") you call this the exploitation/exploration tradeoff: how confident are you that the optimum is near where you are already versus a yet-to-be-discovered new peak. Me, I make a point of adjusting for my own bias towards conservatism, not because I'm unhappy with where I am but because the only way to know you're in a local maximum is to try something unknown.
The other day I went to a friend's apartment in a neighborhood of SF I'd never been to before. He'd bought a big ladder from a hardware store so he could get onto the top of his building, because there we could see a gorgeous view of the Golden Gate. Erinn always chides me for never leaving my own neighborhood, but even here I've yet to go to the restaurant a block from my home. One way of looking at it is that I simply lack the taste for adventure, or that with so many opportunities I'm overwhelmed with the choice. Another way of looking at it is that I have millions of ideas, boxes of tea, stacks of books I have yet to topple. In machine learning ("AI") you call this the exploitation/exploration tradeoff: how confident are you that the optimum is near where you are already versus a yet-to-be-discovered new peak. Me, I make a point of adjusting for my own bias towards conservatism, not because I'm unhappy with where I am but because the only way to know you're in a local maximum is to try something unknown.