From: Neil H. (neuronexmachina@gmail.com)
Date: Mon Nov 13 2006 - 19:31:08 MST
On 11/13/06, Eliezer S. Yudkowsky <sentience@pobox.com> wrote:
> http://www.netflixprize.com/community/viewtopic.php?id=401
>
> 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".)
There's a neat thread on that forum which tells a little bit about who
the people on the leaderboard are:
http://www.netflixprize.com/community/viewtopic.php?id=368
It seems that "NIPS Reject" is a PhD student of Geoff Hinton, a
well-known figure in the neural-networks community. I don't know if
this is the same work, but they published a Science paper a few months
ago, on "Reducing the Dimensionality of Data with Neural Networks":
http://www.cs.toronto.edu/~rsalakhu/papers/science.pdf
http://www.cs.toronto.edu/~rsalakhu/papers/perspective.pdf
http://www.cs.toronto.edu/~rsalakhu/
I actually hadn't seen this paper before -- it's nice to see that
after all these years somebody's managed to tame autoencoder networks
into do something practical. From the end of the accompanying
Perspective article:
"This makes it practical to use much deeper networks than were
previously possible, thus
allowing more complex nonlinear codes to be learned. Although there is
an engineering flavor to much of the paper, this is the first
practical method that results in a completely invertible mapping, so
that new data may be projected into this very low dimensional space.
The hope is that these lower dimensional representations will be
useful for important tasks
such as pattern recognition, transformation, or visualization. Hinton
and Salakhutdinov have already demonstrated some excellent results in
widely varying domains. This is exciting work with many potential
applications in domains of current interest such as biology,
neuroscience, and the study of the Web.
"Recent advances in machine learning have caused some to consider
neural networks obsolete, even dead. This work suggests that such
announcements are premature."
-- Neil
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:57 MDT