Seminar on Computational Learning and Adaptation

 
Some Experiments with Different Learning Techniques

David Bree
University of Manchester, UK
and
CSLI, Stanford University
http://www.cs.man.ac.uk/~dbree

This talk will give an overview of the experience that my students have had over the years with trying to understand and predict daily fluctuations in financial markets. I will begin with a system which learned rules that successfully predicted the price of spot oil up to four days ahead. Then I will look at some systems for predicting short-term fluctuations in foreign exchange rates from the history of a batch of different foreign-exchange rates using:

• backpropagation in multilayer neural networks,
• a battery of support vector machines,
• a genetic algorithm to select the structure of a neural net.

Return to the seminar schedule