View Project

Norwegian AI Directory

New Information Theoretic methods for Intelligent Data Analysis


The goal of the three year IT-IDA postdoctoral project is to develop new machine learning algorithms for intelligent data analysis, such as robust non-linear clustering and classification techniques, based on recently developed concepts in information the ory, Mercer kernel theory and graph spectral theory. Intelligent data analysis deals with the automatic detection of linear and non-linear patterns in data, and is an important research topic in machine learning and artificial intelligence. It also plays an important role in other areas of research, such as e.g. bioinformatics, remote sensing, text categorization, image analysis and web retrieval. Mercer kernel methods are considered the current state-of-the-art of machine learning, enabling classifica tion, clustering, correlation and ranking to be done in a non-linear feature space. In addition, graph spectral methods are also emerging rapidly, and have proved to be very useful, especially in data clustering. The project will focus on theoretical issu es as well as application potential of a new information theoretic approach to machine learning, which is based on recent information theoretic cost functions and non-parametric estimation of probability density functions. This approach has been shown to be closely linked to kernel methods and graph spectral methods, but is in some sense much more intuitive. It also has the advantage of providing more apparent procedures with respect to the selection of system parameters. All together, this provides us ne w tools for a better mathematical understanding, and for developing better algorithms. It even opens avenues for fusing together the best parts of the above mentioned related research areas, which to date have been considered separate fields. The project will lead to new and powerful information theoretic machine learning algorithms for intelligent data analysis, which will be aimed at operational use in applied areas such as remote sensing and bioinformatics.

Project leader: Robert Jenssen

Started: 2006

Ends: 2009

Category: Universiteter

Sector: UoH-sektor

Budget: 2493000

Institution: Institutt for fysikk og teknologi

Address: Tromsø