|Universiteit Faculteit FNWI English version||3 maart 2009, e-mail|
Woensdag 11 maart 2009
AbstractArtificial intelligence aims to capture intelligent behavior in machines. It maybe quite difficult to state what intelligence exactly is. Instead, I will restrict myself to the more technical area of machine learning and argue that there are some necessary aspects that machines must possess, such as advanced pattern recognition, associative memory, learning and planning.
To describe these problems formally in an environment that is noisy and where the animal or agent has only access to partial information, one commonly uses probability theory. This approach offers a rich and powerful modeling framework and I will give an example of how probability methods are used for planning. However, probabilistic methods have the disadvantage that even modest size computations become intractable, either in terms of memory requirements or in terms of computation time or both.
I will introduce some generic approaches to efficiently approximate these computations using methods from statistical physics. If time permits, I will illustrate this on a medical expert system.