The field of artificial intelligence (AI) is undergoing a revival, spurred by probabilistic programming that merges classic AI's logical principles with the power of statistics and probability. The key to probabilistic reasoning is a Bayesian network, a model comprised of various random factors, each with a probability distribution that is dependent on every other factor. Given the value of one or more factors, the network enables the inference of all the other factors' likely values.
The development of algorithms for Bayesian networks that could use and learn from existing data began in the mid 1990s, and these new algorithms were capable of learning models of greater complexity and accuracy from much less data, unlike artificial neural networks. Still, Bayesian networks are insufficient for modern AI challenges on their own because they are incapable of building arbitrarily complex constructions out of simple components, which is where the incorporation of logic comes in. At the vanguard of probabilistic programming are computer languages that use both Bayesian networks and logic programming.
In addition to the development of fast and flexible inference algorithms, researchers face the challenge of improving AI systems' learning ability, whether from existing data or from the real world using sensors.
From New Scientist
View Full Article
Abstracts Copyright © 2011 Information Inc., Bethesda, Maryland, USA