This article is more than 1 year old
Intel to offer Bayesian computer learning code
Boosting demand for faster chips
Intel is planning to give away a library of software designed to improve a computer's ability to learn, NewScientist.com reports.
It will also, of course, make it more desirable that that computer has a more powerful processors.
The code uses so called Bayesian networks, a combination of statistical principles, including Bayes Theorem, hence the name. Discovered among the papers of the English mathematical the Reverend Thomas Bayes after his death in 1761, the theorem lets you work out how the likelihood of an event changes as you gain more information about the event.
Bayesian statistics is a bit of a scientific flavour of the month right now, largely by forming the basis of the some of the more effective anti-spam filters. Software based on Bayes Theorem, having been 'trained' to spot words common to span, can then go off and use that information to work out the likelihood that an incoming message containing those words is also spam. It's not perfect - some spam gets through by coming in under the spam likelihood threshold - but no real messages are mis-diagnosed as unwanted advertising.
The team behind Intel's Bayesian code library sees broader uses of its software, from data-mining to computer vision and decision-making systems - indeed, any application where the computer has to make a 'guess' based on a set of rules.
But if Bayesian networks are so smart, why are they only now coming to prominence. Because they are computationally intensive. Suddenly Intel's motivation is revealed. We're sure the development team are driven by the pure computational science of it all, but their employer clearly benefits by any application that drives the need for faster chips.
Intel hopes to release the library at the Neural Information Processing Systems 2003 conference to be held in Vancouver next December. ®