Empirical Methods for Artificial Intelligence
Slawek Nowaczyk, AI@CS, LU
AI seminar, November 16th, 2006


Abstract:

This seminar is based on a tutorial with the same title that I attended
during this year's AAAI conference. The tutorial was given by Paul R.
Cohen, from University of Massachusetts Amherst, and I found it very
interesting and enlightening. Even though most of the things included
are not exactly new, it still offered a refreshing view on the practical
aspects of experiment design and on empirical data analysis.

Some of the things covered in the tutorial that I would like to talk
about in my seminar are the following: Understanding and explaining
variability; The relationship between theoretical and empirical science;
Exploratory data analysis; Data transformations; Experiment design; The
purposes of experiments; Hypothesis testing and estimation; Comparing
things; Predicting things.

This seminar is not going to be an introduction to statistics (even
though statistics is omnipresent in empirical work), since I assume all
of the listeners are familiar with it. Rather, I will try to present
some "tips and tricks", as well as pitfalls, that are not immediately
obvious.