PhD Seminar Course on

Multiple Classifier Systems

Cagliari, June 25-27, 2008

Instructor: Josef Kittler
Centre for Vision, Speech and Signal Processing, University of Surrey (UK)
Duration: 8 hours
Schedule: Wednesday 25, 15-18
Thursday 26, 10-13
Friday 27, 10-12
Venue: Aula Y (June 25 and 26), Aula X (June 27)
Topics: The performance of pattern recognition systems can be improved by drawing on the consensus of opinions provided by multiple pattern recognition experts. The series of six lectures will introduce the problem of designing multiple classifier systems and discuss a generic architecture, and fusion methodology supporting both intra-modal and multimodal multiple classifier system paradigms. The effectiveness of multiple classifier systems will be illustrated on several studies in personal identity authentication using biometrics. The lecture course will cover the following topics:
1. Introduction (pattern recognition problem, PR system, statistical decision rules, multiple classifier paradigm, multiple classifier system architectures, score normalisation)
2. Multimodal expert fusion (problem formulation, simple fusion rules and their relationship, error sensitivity, trained rules)
3. Coping with estimation errors (Bayesian error probability, error distribution, actual error probability, intramodal fusion architectures, diversification and marginalisation) 4. Margin manipulation approaches (error correcting output coding approaches, serial multiple classifier systems)
5. Classifier selection
6. Quality based fusion of multiple experts
Assessment: To decide
Organizer: Fabio Roli
Dep. of Electrical and Electronic Engineering
University of Cagliari, Italy