Knowledge
Discovery through Machine Learning:
John Moody
Computational Finance Program
Oregon Graduate Institute
moody@cse.ogi.edu
Machine Learning algorithms have proven
to be powerful tools for finding patterns in data and discovering solutions to
difficult problems in a wide range of fields. In this presentation, I'll
survey three styles of machine learning, focusing on key algorithms, and
explaining why they are innovative and powerful. These include supervised
learning (w/ multi-layer perceptrons), unsupervised learning (e.g. independent
component analysis) and reinforcement learning. Benefits to these methods
include bypassing Bellman's curse of dimensionality, discovering subtle or
higher order statistical structures in data, and developing efficient
strategies for making real-world decisions.
I'll describe application examples from
several fields, time permitting, including computer games, signal processing,
forecasting the U.S. economy and trading the global financial markets.