Two Ways of Looking at Machine Learning

Abstract

On the surface, machine learning can seem to mean many different things. Some techniques learn to separate numerical data into different groupings. Others learn to classify photographs or paragraphs of text according to their contents. Still other algorithms find plans of action, for instance learning to find paths through mazes, or control robotic arms. This talk will look at two different approaches to machine learning: classification of image data on the one hand, and path planning on the other. While these can look quite different, there are in fact many similarities to do with the mathematics that makes each possible. We will show the results of different learning algorithms, applied to different data sources, and outline just enough of the math behind it all to give a sense of how it all really works. The talk is designed to be understood by those who have not studied machine learning or AI before, and will hopefully have something new even for those who have.

Bio

Marty Allen received his doctorate in computer science from the University of Massachusetts Amherst in 2009, where he worked in artificial intelligence and theoretical computing areas. While his undergraduate and first graduate degrees were in philosophy, he got interested in CS as an opportunity to do collaborative and interdisciplinary research, recognizing the ways in which computing can be applied in subjects spanning both the sciences and the arts. For most of the past decade, he has worked as an educator, mainly at the University of Wisconsin-La Crosse, where he is currently an associate professor in the Computer Science department. His recent research involves multiagent reinforcement learning, where teams of individuals try to learn how to find cooperative solutions to complex problems, and the application of simulation and machine learning to complex biological systems.