Foresights
Bayesian machine learning
What is it?
Though ‘know thyself’ was a goal worthy enough to be inscribed on the Delphic Oracle, it now seems machines may know us even better than we know ourselves. Using Bayesian machine learning, spam filters can reduce unwanted e-mail by more than 99 percent – a skill they pick up by watching our behavior.
About 300 years ago Thomas Bayes, a non-conformist minister, invented a technique that was a heresy against traditional statistics. Nurtured in the broad church of artificial intelligence by devotees of developmental psychology, computing and cognitive science, the heresy rose again in Bayesian machine learning.
Bayesian networks (BNs) are the soul of Bayesian machine learning and are a way of managing uncertainty and complexity by learning from experience. They infer cause and effect (rather than having to be told it) by modifying an initial theory using evidence and a fixed set of rules based on Bayes’ theorem. Using BNs, machines ‘learn’ about the real world in which cause and effect are often uncertain.
|
