Current music:

http://www.tcpmusic.com/ -- Red Horizon

principal component analysis for fun and profit
The hard part is knowing everything I'm writing will be worthless in three months.
knowing that the next version I write will be ten times better than today's nonsense.
knowing that I could save so much time just by writing the next version.
knowing that the only way to write the next version is to write today's first.
knowing that the mistakes of today will be obvious only after I've made them.

"...a market economy is essentially a genetic algorithm for solving resource allocation problems, and while the Soviets were grappling with abacuses and five year plans, the western economies were automating their genetic algorithms and adding all sorts of weird feedback loops in the form of derivatives and options trades. And the western economies were also moving into a new, post-industrial phase. Efficiency killed off the first industrial revolution's economy, so that from having 50% of the population working in factories, we have to find service industry jobs for them. And this change happened so damned fast that the GOSPLAN five year feedback cycle was stuffed." --Stross

The cool part is that given the available data, an ideally intelligent actor should act in a way that is totally predictable.

We have to leave the city, immediately. And avoid the authorities.
Can I stop by my house?
Negative. The T-1000 will definitely try to reacquire you there.
You sure?
[shrugs] I would.

And the awful part is that I have no idea what that way is. The cute part is watching two AI killing machines use empathy to shrink each other's heads through the movie. Too bad the bad guy was merely stronger. I want to know how to beat someone who's /smarter/. How do you get inside the head of an enemy who's like that?

Might an ideally intelligent actor have to avoid doing the Best Thing because doing so would reveal too much about what it knows to others like itself? In fact, if you assume the existence of good competition you have to sometimes screw up in order to hide your true knowledge from They Who Pose a Threat. Could you screw up in an organized fashion, so as to project a coherent-but-false impression about your information and lure your enemy into a compromising position? Naturally, you must assume your enemy will expect you to do just that, so the illusion will have to be multi-layered, at which point you're pretending to not know so much that you've sacrificed most of your abilities in order to appear weaker than you are (with the occasional random amazing stroke of luck that wins the day, of course).

http://www.infinityplus.co.uk/stories/colderwar.htm -- A Colder War, Charles Stross, a good bedtime novella.

http://www.cs.tut.fi/~lasip/ -- "Local Approximations in Signal and Image Processing (LASIP) is a project dedicated to investigations in a wide class of novel efficient adaptive signal processing..." -- using, I may add, nonparametric scale-adaptive kernel regression. See also http://en.wikipedia.org/wiki/Kernel_regression and http://en.wikipedia.org/wiki/Kernel_smoothing and http://en.wikipedia.org/wiki/Kernel_density_estimation which is called (in Matlab) ksdensity.

...using non-Bayesian methods such as minimax, the loss function should be based on the idea of regret, i.e. the loss associated with a decision should be the difference between the consequences of the best decision that could have been taken had the underlying circumstances been known and the decision that was in fact taken before they were known...

In probability and statistics, density estimation is the construction of an estimate, based on observed data, of an unobservable underlying probability density function. The unobservable density function is thought of as the density according to which a large population is distributed; the data are usually thought of as a random sample from that population.

A kernel is a weighting function used in non-parametric estimation techniques. Kernels are used in kernel density estimation to estimate random variables' density functions, or in kernel regression to estimate the conditional expectation of a random variable. Kernels are also used in time-series, in the use of the periodogram to estimate the spectral density. An additional use is in the estimation of a time-varying intensity for a point process.

...used in many fields, including machine learning, data mining, pattern recognition, image analysis and bioinformatics. The computational task of classifying the data set into k clusters is often referred to as k-clustering.
--Wikipedia

Note to self: kernel density estimation is easy in one dimension because there's only one scale factor. N-space will involve n scale factors (unless you want some axes to be undersmoothed while others are oversmoothed).

there is no past, only memory
there is no future, only hope

http://mdp-toolkit.sourceforge.net/ -- "MDP has been written in the context of theoretical research in neuroscience, but it has been designed to be helpful in any context where trainable data processing algorithms are used. MDP has been originally written by Pietro Berkes and Tiziano Zito at the Institute for Theoretical Biology of the Humboldt University, Berlin in 2003."