March 11th, 2009


machine learning Nargery [1]

So I saw a talk today at the lab that had an interesting machine-learning twist. In one of the usual cases, we are given n apples and m non-apples, and we build a system that learns how to classify "is-it-an-apple" from these n+m fruit.

In these experiments discussed today, by contrast, we are given n apples, and m other fruit -- mostly non-apples, but some of them are apples and some not. Correspondingly, the goal of future classification is not "is-it-an-apple" but "get-me-six-apples".

As it turns out, this restructuring is an interesting, and different, problem, particularly because there are positive examples (apples) in the second (m) class, making them not quite a "negative" class.

I'm kinda disappointed that both Dorkbot SF and Noisebridge Machine Learning night are tonight; the latter group might be a place to talk about this a little further.

1 nargery is a term I learned from marnanel: "Not A Real Gentleman" refers to people who are actually (and pejoratively, see firinel's comment) interested in their work; he reclaims the term. Thus it is roughly equivalent to "nerdery".
2okay, the problem discussed today wasn't about fruit-sorting. But I don't want to get into that part.