Tema : "Issues and Challenges in Multimodal Biometrics"

 

Abstract:

This course aim to arm the participants by hand-crafing a Bayesian classifier using both generative and discriminative approaches. Advanced topics such as dealing with missing values and Bayesian network are also covered. These techniques enable one to solve many machine-learning problems where training data is an integral part of the system design. Practical applications are illustrated via multimodal information fusion in biometrics.

 

Lectures Scheduling:

16 Oct 10am-12pm: Lecture 1: Decision theory, generative and discriminative classifiers, density estimation, controlling classifier complexity (cross-validation, regularization),

19 Oct 10am-12pm: Lecture 2: Naive Bayes, Dealing with missing values, Bayesian networks

21 Oct 10am-12pm: Lecture 3: Classification combination, Error prediction in multimodal information fusion, Quality-based fusion

23 Oct 10am-12pm: Lecture 4: User-specific score normalization and fusion

 

Lectures Location:  

Breve CV del Relatore: Dr. Norman Poh received a PhD (2006) in Computer Science from the Swiss Federal Institute of technology in Lausanne (EPFL), Switzerland. He then worked with Prof. Josef Kittler as a post-doctoral researcher at CVSSP, University of Surrey (UniS). Currently, he is a research fellow as well as one of the work package leaders in the EU-funded Mobile biometry project. His areas of expertise are multiple classifier systems, pattern recognition and biometrics. He published in excess of 50 scientific papers in these areas and is the main author of three best paper awards: Int'l Conf. on Audio-visual Biometric Person Authentication - AVBPA 2005, Int'l Conf. on Biometrics - ICB 2009, and Pattern Recognition Journal 2006, as well as a recipient of two personal (post-doctoral level) research fellowships (2006 and 2008-2009,

respectively) from the Swiss National Science Foundations.