Greg Detre
17:36 on Wednesday, 07 February, 2001
Dr Colin Campbell, Dept of Eng Maths, Bristol Uni
support vector machines � learning machines
classification, regression, novelty detection etc.
can derive a NN, RBF net, polynomial classifier from them
it�s general in the problems you can tackle
SVMs
fewer choices to be made
convex curve, as opp to back propagation net where there are sub-optimal solutions
kernel
kernel parameter
statistical learning theory � simplest task is binary classification, separating hyperplane is maximally distant from the two clusters
utilise support vectors to help separate out the two clusters