 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
   
|
Division of Algorithmic Data Analysis
The Research Division of Algorithmic Data Analysis at Rolf Nevanlinna Institute, University of Helsinki, developed computational methods for modern data analysis and applied them on challenging problems in science and engineering. The research covered a spectrum of methods from algorithms for knowledge discovery and data mining to Bayesian statistical tools and models. The division worked in a close cooperation with other research divisions of the Institute, with the Department of Computer Science, and with other departments of the University. In 1999 the Division consisted of Dr. Hannu TT Toivonen (Division Head), Jouni Seppänen, Petteri Sevón, M.Sc., Kari Vasko, M.Sc., and Dr. Bob O'Hara (Visiting Researcher).
Data mining and knowledge discovery Data mining methods can be used to analyze large databases and to discover regularities and hidden relationships in them. The most recent research topic of the Division, in collaboration with the Division of Biometry and the Finnish Genome Center (Prof. Juha Kere), was the development of gene localization methods, based on patterns in genetic marker data. Data mining tasks studied in the Division include the discovery of functional dependencies in relational databases, efficient and flexible similarity comparisons in time series databases, and the use of taxonomical information in pattern extraction. Algorithms and tools developed by members of the group have been used in remote monitoring tasks in the industry. Some of these studies have been carried out in collaboration with Prof. Heikki Mannila, Dr. Mika Klemettinen, or Dr. Juha Kärkkäinen, (Department of Computer Science), and Finnish companies. A recurrent research theme for the Division has been the discovery of frequently occurring logical patterns or structural subcomponents in large databases. An interesting application is the discovery of relationships between structural properties of chemical constituents and their functions in living organisms, such as carcinogenecity . This line of research has benefited from collaboration with Dr. Luc Dehaspe (Katholieke Universiteit Leuven, Belgium).
Computational tools for Bayesian inference An important topic of the group was the development of tools for MCMC (Markov chain Monte Carlo) approximation of posterior distributions of hierarchical Bayesian models. Such models offer a powerful framework for modeling statistically complex real-world phenomena, but the lack of computational tools has hindered their practical use. The research group has developed, in cooperation with the Department of Computer Science, a publicly available software package, Bassist, that automates the tedious process of developing MCMC programs. In a joint effort with Dr. Atte Korhola (Department of Ecology and Systematics), the group has developed biologically viable Bayesian models for producing reliable reconstructions of past climate. Another research application of Bassist was in the statistical modeling of metapopulations. This research is being carried out jointly with Dr. Bob O'Hara and Prof. Ilkka Hanski (Department of Ecology and Systematics) and Prof. Elja Arjas (Division of Biometry). Further links:
|
  |
