From: Eliezer S. Yudkowsky (email@example.com)
Date: Mon Nov 13 2006 - 14:00:46 MST
This is a forum devoted to the Netflix Prize, $1 million for producing a
collaborative filtering algorithm 10% better than Netflix's. The
current leading contenders are edging up on 5% better than Netflix's
algorithm, corresponding to a root mean squared error of .90. (I
haven't taken a potshot at this problem yet, but it's quite interesting
to see how things go. Right now, the current leading algorithm, beating
out many serious contenders, is apparently one that was rejected from
the NIPS conference as uninteresting. Hence the name, "NIPS Reject".)
Anyway, the first two posts from the above thread are just... beautiful.
People sure are fast to give up, aren't they?
The leaderboard has stalled at .90. It seems that super-powered
computers combined with super-descriptive algorithms may be reaching
their maximum predictive power.
There is a historical parallel. Imagine that we had given 100,000,000
astronomical observations to astronomer Claudius Ptolemy around 130 A.D.
He would have come up with an astrolabe able to describe the past
motions of the known planets very accurately. But it would not be so
capable of predicting the future motions of those planets. The reason is
that his Earth-centered theory of planetary motion was deficient.
It took Copernicus, then Newton, then Einstein, to develop better
theories of planetary motion. Copernicus theorized that the Sun is the
center of planetary circles. This is more substantively accurate than
that the Earth is at the center, but Copernicus's circle-based
predictions of the locations of the planets were less accurate than
Ptolemy's, as were also Newton's first ellipse-based predictions.
So now we have the same situation with the Netflix data. It seems that
the super-descriptive algorithms are based on deficient psychometric
theory. (Psychometry is the branch of psychology that deals with these
types of data). In the past 100 years, several psychometric theories
have been developed, but these appear to be at the Copernicus or early
Newton stages of development. Applying them to the Netflix data does not
yield as good results as the Ptolemaic super-descriptive approach.
It seems that we need a psychometric Newton, or even an Einstein, to
raise the prediction of ratings to a higher order of accuracy.
Dude, with all due respect it's been only several weeks since the start
of the contest. Drawing analogies with Copernicus is all good and nice
but there is no way the winning algorithm doesn't exist today.
-- Eliezer S. Yudkowsky http://intelligence.org/ Research Fellow, Singularity Institute for Artificial Intelligence
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:57 MDT