The International Workshop on Multiple Classifier Systems is a series of conferences whose goal is the development of theories, algorithms,
and applications of ensemble machine learning methods. This series, started in the year 2000, was spawned by the theoretical and empirical successes of ensemble methods especially in the 1990s. The series has served as a forum for the various groups working in this area to come together and help each other keep up with this active area of research. The fifth in this series of conferences was held last June in beautiful, sunny Sardinia. The beautiful beaches, casual atmosphere, and delicious Italian food led to educational and energetic talks and discussions.

 

This conference contained the usual excellent mix of theory, algorithms, and applications of ensemble methods. The conference included a panel discussion, organized by Prof. Joydeep Ghosh and Dr. Nikunj Oza. The first of the discussion's two components was a story telling competition in which people shared anecdotes of successes or failures of particular ensemble approaches. The second component was a discussion of how to carry the field forward, including topics to de-emphasize and topics that need greater attention. The topics that people felt needed greater attention can be summarized as applying ensemble methods to modern data mining problems. These include problems that have changing environments, data distributed across many computers, complex data forms (e.g., text), and/or other characteristics that make the use of traditional machine learning algorithms difficult.

 

This theme also appeared in the first two invited talks. The first talk, "Classifier Ensembles for Changing Environments," by Lucy Kuncheva, neatly summarized the work that has been done so far that goes in the direction of working on problems with changing environments, and discussed the gaps that need to be filled. Nagi Rao's talk, "A Generic Sensor Fusion Problem: Classification and Function Estimation," discussed sensor fusion, which is becoming an increasingly important problem with the growth of the area of Integrated System Health Management (ISHM). The third talk, "Security and Dependability in the Ambient Intelligence Space: Challenges and Opportunities for European R&D" by Andrea Servida also discussed a problem---security---whose importance is greatly increasing in our modern world.

 

The talks and discussions of MCS 2004 gave the attendees an excellent overview of the work done within the last year in the field of ensemble learning methods as well as new problems to focus on. Next year's conference should reveal important progress in these new directions as well as interesting extensions to active topics. The next MCS will be held in the USA (Monterey, California, www.monterey.org) for the first time, but will still provide as much time in the sun as last year's conference. The dates will be June 13-15, 2005. Please watch www.diee.unica.it/mcs over the next few months for further developments.

Workshop Report:  MCS 2004

General Chairs:

H. Fujisawa

Guy Lorette

Return to Home Page

For more information:

MCS Conference Proceedings