For IJCNN 2006

July 16 - 21, 2006, Vancouver, Canada

Ensemble in the context of machine learning is viewed as a paradigm in which multiple individual models (e.g. classifiers or predictors) are trained to learn a given task from data and then work collectively to produce an overall (hopefully better) decision for an unseen instance through a decision fusion strategy. Ensemble approach has been popularised particularly in recently years primarily due to reasons as follows:

• the complexity and the sheer quantity, yet ever increasing rapidly, of various real world data;
• the unsatisfactory performance of individual techniques when dealing with real world problems;
• the ever improving powerful computing hardware and software facilities make construction of ensembles easier, cheaper and quicker. Particularly in inductive learning, inductive models (e.g. decision trees, neural networks) can be conveniently trained in large quantity as model candidates for constructing an ensemble.

This special session aims to provide an opportunity to bring researchers who are interested in this topic together to review and evaluate the latest development on ensemble methods and the effectiveness of diversity measures, and to exchange and discuss new ideas for building more effective ensembles.

• Ensemble methodology, strategies and techniques,
• Diversity definition and measurement,
• Strategies and techniques for generating diverse models (classifiers/predictors, any other models),
• Relationships between diversity between individual models and ensemble's accuracy,
• Evaluation of ensemble performance and comparisons with other approaches,
• Decision fusion strategies,
• Ensemble Implementation(software development),
• Real world applications and (both positive and negative) experiences.

Submission

• Notification: March 15, 2006
• Camera ready: April 15, 2006

Special Session Organiser

Special Session Technical Committee