Dealing with Indefinite Representations in Pattern Recognition
Faculty of Electrical Engineering, Mathematics and Computer Science
Delft University of Technology, The Netherlands
Friday, March 14, 2008
Talk: 10:00 am - 11:00 am
Host: Anil Jain
The two main steps in the construction of a pattern recognition system are representation and generalization. The first defines the playing ground for the latter. Traditionally, features are used for building a representation. Other vector representations may be considered as well in situations where it is difficult to define good features from the available background knowledge of the application.
A review will be presented of the possibilities of using dissimilarities, similarities and kernels. We will focus on the problem non-Euclidean data and explain why it is frequently encountered in practice. Various ways to train classifiers for such data will be discussed and compared, including approaches to 'correct' non-Euclidean data such that it can be mapped into a Euclidean space.
Robert P.W. Duin studied applied physics at Delft University of Technology in the Netherlands. In 1978 he received the Ph.D. degree for a thesis on the accuracy of statistical pattern recognizers. In his research he included various aspects of the automatic interpretation of measurements, learning systems and classifiers. Between 1980 and 1990 he studied and developed hardware architectures and software configurations for interactive image analysis. After this period his interest was redirected via neural networks to pattern recognition.
At this moment he is an associate professor of the Faculty of Electrical Engineering, Mathematics and Computer Science of Delft University of Technology. His present research is in the design, evaluation and application of algorithms that learn from examples. This includes neural network classifiers, support vector machines, classifier combining strategies and one-class classifiers. Especially complexity issues and the learning behavior of trainable systems receives much interest. Recently he started to investigate alternative object representations for classification and he became thereby interested in dissimilarity based pattern recognition, trainable similarities and the handling of non-Euclidean data. He expects that this will contribute to the unification of learning from structure and learning from statistics.
The pattern recognition research team headed by Robert Duin studies many industrial and medical applications. They are thereby interested in pattern recognition system design, the handling of ill-sampled problems and in varying costs and prior probabilities. A series of pattern recognition courses for industry has been set up by this team. The software (PRTools) is public available for research purposes.
Robert Duin is a former associate editor of IEEE Transactions on Pattern Analysis and Machine Intelligence. Presently he is an advisory editor of Pattern Recognition Letters. He is a member of the IEEE, and a fellow of the IAPR. In August 2006 he received the Pierre Devijver Award for his contributions to statistical pattern recognition.