ICML-2002 Tutorials



Monday, 8 July, 10am

Bayesian Kernel Methods

Dr. Alexander Johannes Smola, The Australian National University, Canberra, Alex.Smola@anu.edu.au

The tutorial will introduce Gaussian Processes both for Classifcation and Regression. This includes a brief presentation of covariance functions, their connection
to Support Vector Kernels, and an overview over recent optimization methods for Gaussian Processes.

Target Audience: Novices and researchers more advanced in the knowledge of Gaussian Processes will benefit from the presentation. While being self contained, i.e., without requiring much further knowledge than basic calculus and linear algebra, the presentation will advance to state of the art results in optimization and adaptive inference. This means that the course will cater for Graduate Students and senior researchers alike. In particular, I will not assume knowledge beyond undergraduate mathematics (see Prerequisites for further detail).

Expected Knowledge Gain: a working knowledge in Gaussian Processes which will enable the audience to apply Bayesian inference methods in their research without much further training.

Prerequisites: Nothing beyond undergraduate knowledge in mathematics is expected. More specifically, I assume:

See http://mlg.anu.edu.au/~smola/summer2002 for details.

Monday, 8 July,  2pm

Inside WEKA -- and Beyond the Book

Ian H. Witten, Computer Science, University of Waikato, NZ,  ihw@cs.waikato.ac.nz
Eibe Frank, Computer Science, University of Waikato, NZ, eibe@cs.waikato.ac.nz
Bernhard Pfahringer, Computer Science, University of Waikato, NZ,  bernhard@cs.waikato.ac.nz
Mark Hall, Computer Science, University of Waikato, NZ,  mhall@cs.waikato.ac.nz

Weka is an open-source Weka machine learning workbench, implemented in Java, that incorporates many popular algorithms and is widely used for practical work in machine learning.  This tutorial describes and demonstrates the many recent developments that have been made in the Weka system.  It also looks inside Weka and sketches its inner workings for people who want to extend it with their own machine learning implementations and make them available to the community by contributing to this open-source project.The goal is to empower attendees to increase the productivity of their machine learning research and application development by making best use of the Weka
workbench, and to share their efforts with others by contributing them to the ML community.  The tutorial is aimed at people who want to know about advanced features of Weka, and also at those who want to work within the system at a programming level.

This tutorial is *not* intended as a comprehensive introduction to Weka: attendees are presumed to have some familiarity with it already.  Neither does it reveal any secrets that are not in the current version of Weka: if you have fully explored the features in the latest distribution you do not need to attend the tutorial.

Attendees are expected to have:

    1.    Basic knowledge of ML algorithms and methodology
    2.    Some familiarity with Weka
    3.    Some programming experience in Java.

(The book "Data mining" by Witten and Frank covers all three at an appropriate level).
There will be a 2-hour lab session after the tutorial for those who want to follow up with some practical work.  Computers (Linux) will be available, or you can bring along your laptop (Windows or Linux) and we will help you install the latest version of Weka from a CD-ROM.  We will provide exercises for you to work on; alternatively you are encouraged to bring their own data files and use Weka on them instead.  Tutorial help will be available throughout the lab session


Tuesday, 9 July,  2pm

Introduction to Minimum Length Encoding Inference

Dr David Dowe,Monash University, Australia, dld@cs.monash.edu.au

The tutorial will be on Minimum Length Encoding, encompassing both Minimum Message Length (MML) and Minimum Description Length (MDL) inductive inference, topics central to the 1999 special issue of the Computer Journal on Kolmogorov complexity (vol. 42, no. 4, 1999).  This information-theoretic approach bridges many fields, and is yielding state-of-the-art solutions to at least many problems in machine learning, statistics, econometrics and ``data mining''.  It has applications right across the sciences.

This work is information-theoretic in nature, with a broad range of applications in machine learning, statistics, knowledge discovery and data mining. We discuss statistical parameter estimation and mixture modelling (or clustering) of continuous, discrete and circular data. We also discuss learning decision trees and decision graphs, both with standard multinomial leaf distributions and with more complicated models. We further discuss MML solutions of either cut-point problems or polynomial regression; and, if time permits, possibly Support Vector Machines (SVMs), causal networks and finite state machines (Hidden Markov Models, HMMs) or other problems.
The target audience is academics, machine learning and data mining  practitioners and consultants, and/or others with at least first year university education in at least one of
mathematics, statistics, econometrics or electrical engineering. The audience will learn introductory fundamentals of MML inference, and the many state-of-the art success of MML in statistics, machine learning and hybrid problems. The audience will also see some applications of MML to real-world data.