http://phoenix.fit.edu/~georgio/

AG 1, AG 3, AG 5, SWS, AG 4, RG1, MMCI   Info

AG Audience

English

30 Minutes

Saarbrücken

Kernel-based methods have been an intensely researched topic in Machine Learning (ML) over the last ten years. Their popularity stems first from their ability to map data from their native space into a feature space, in which a ML problem may be easier to solve. The map itself is implicitly defined via the choice of kernel. Secondly, they allow the application of many traditional ML approaches to problems dealing with data that are not purely numeric in nature by appropriate constructing kernels for this type of data.

Additionally, Metric Multi-Dimensional Scaling (metric MDS) has been a technique for visualizing data, when only pairwise similarities or dissimilarities between data are available. This type of proximity information is then depicted as pairwise distances in a low-dimensional space, where relationships between data can be visually assessed. Such ability finds important applications, in engineering, social sciences, etc.

This talk features two parts. The first part provides a high-level overview of fundamental concepts of kernel-based learning, such as Reproducing Kernel Hilbert Spaces, positive definite kernels, and the role of these kernels in Machine Learning. The second part provides some background on MDS and showcases the speaker’s past and current efforts in developing a general kernel-based metric MDS approach along with some application examples.

Tobias Friedrich

--email hidden

passcode not visible

logged in users only