Research projects
Information management
Independent component analysis and its extensions
Period: 4/2003-12/2005
Researchers: Aapo Hyvärinen, Patrik Hoyer, Shohei Shimizu, Antti Kerminen, Urs Köster, Jukka Perkiö
Funding: HIIT/BRU, The Academy of Finland, international foundations
The analysis of independent components is a model for data analysis where multi-dimensional measurement data is presented as a linear combination of hidden, statistically independent components.
We are developing new and more efficient variants of the method, utilising e.g. time behaviour and interdependencies. The project also studies how the methodology can be applied to causal analysis. A further object of study is the connection between this method and models of multi-dimensional data analysis, such as positive matrix factorisation, models of structure equations and graphical models.
Advanced data analysis in vision research
Period: 01/2004 - 12/2006
Researchers: Ilmari Kurki, Aapo Hyvärinen
Funding: The Academy of Finland
We develop new ways of analysing data measured on the performance of the human vision system. Our approach is based on a recently developed experimental paradigm, so-called classification images. This is a co-operation with the Helsinki University Department of Psychology.
Statistical modelling of image and video data
Period: 04/03 - 12/2007
Researchers: Aapo Hyvärinen, Jarmo Hurri, Urs Köster, Jussi Lindgren
Funding: HIIT/BRU, The Academy of Finland, HeCSE, international foundation
We develop new statistical models of image and video data. The models are useful for both human-vision research and computer vision and image processing. In 2005, we mainly developed models of statistical characteristics of nonlinear features.
Application of Probabilistic Inductive Logic Programming II (APrIL II)
Period:1/2004 – 12/2006
Researchers: Aristides Gionis, Heikki Mannila, Taneli Mielikäinen, Evimaria Terzi, Panayiotis Tsaparas
Funding: EU
Probabilistic inductive logic programming combines probabilistic modelling and inductive logic programming to provide a general framework for learning probabilistic logical models from structured data.
The goal of the APrIL II project is to study the theoretical basis of probabilistic inductive logic programming, develop efficient computational models for the estimation of the structures and parameters of probabilistic logical models, and to apply these models to practical modelling problems, especially in the field of bioinformatics. Researchers from the Albert-Ludwig in Freiburg, Imperial College of Science, Technology and Medicine, INRIA Rocquencourt, Helsinki Institute for Information Technology and the University of Florence participate in the project. In Helsinki, the research is focused on algorithmics for probabilistic inductive logic programming, especially algorithmics for segmentation problems.
Context recognition by user situation data analysis (Context)
Period: 1/2003-12/2005
Researchers: Mika Raento, Kari Laasonen, Renaud Petit, Hannu Toivonen
Funding: The Academy of Finland
This project studies the characterisation and analysis of user contexts, and the adaptation of context data in proactive computing. The research issues of the project include the users’ concepts of their contexts, automatic conclusions based on context information, as well as how to present context data to the user. The project is a co-operation with the User Experience group at the Applied Research Unit at HIIT.
The project developed the ContextPhone software, which collects, saves and relays context information in normal S60 mobile phones. It can also automatically annotate the pictures taken with camera phones with context information and move them e.g. to a website. With the help of this software, the project has studied the effect of relaying context information on user communications and developed methods for refining cell-based positioning data into a more graphic form. ContextPhone has been used in research at Berkeley, MIT, and the University of Art and Design Helsinki, among others.
Computational methods for the analysis of palaentological data
Period: 01/2005- 12/2008
Researchers: Heikki Mannila, Ella Bingham, Hannes Heikinheimo, Kai Puolamäki, Antti Ukkonen
Funding: The Academy of Finland, HIIT/BRU
The project develops computational methods for the analysis of palaentological and other ecological data. The project has developed new dating methods based on spectral ordering and MCMC methods. A method using MCMC and the so-called Bernoulli model were used to discover errors in the data.
The hierarchical structure of modern mammal data was studied with the help of distance measurements based on distribution, and the automatic learning of the hierarchy was analysed. The project is a cooperation with the Department of Geology at the University of Helsinki.
Mobile and multilingual maintenance man - (4M)
Period: 08/2003-07/2007
Researchers: Reeta Kuuskoski, Helena Ahonen-Myka, Damien Beaudrey, Antoine Doucet.
Funding: Co-operation project, other partners: UH/language technology, UH/Department of Translation Studies, HUT/BIT research institute, HUT/SoberIT/usability group, HUT Knowledge engineering, VTT Information technology, TEKES, Fujitsu Services, Nokia Research Center, Nokia Business Infrastructure, Pasanet/Lingsoft, RAY, Wärtsilä
The project works on developing a knowledge support system for appliance maintenance. The system includes a dialogue component that uses natural language. The knowledge about appliances and problems is recorded in ontologies. It also searches for instructions in the documentation for the appliances if the answer cannot be found in the ontologies.
The first prototype integrating all components was completed in September. We were responsible for the information retrieval component that uses the ontologies. In addition, we developed document-structuring methods, such as methods for recognising user instructions in the documentation. The structuring of documents makes real-time problem-solving more efficient.
New computational techniques for analysing the structural and functional landscape of the mammalian genomes (CompGenome)
Period: 01/2004- 12/2007
Researchers: Heikki Mannila, Aristides Gionis, Niina Haiminen, Jaana Wessman, Mikko Koivisto, Jussi Kollin, Kimmo Palin, Panayiotis Tsaparas
Funding: The Academy of Finland
The project studies species-specific and inter-species genetic and functional varieties. The goalis to understand multifactorial diseases, among others. The biological themes include haplotype structures, large-scale genetic variations, phenotype clustering, and gene expression. The main computational themes are probabilistic modelling and MCMC methods, data mining and pattern discovery, and combinatorial algorithms. The project, a cooperation with the Finnish Genome Centre, the National Public Health Institute and Karolinska Institutet in Stockholm, belongs to the Academy of Finland’s SYSBIO programme.
During its second year, 2005, the project continued segmenting genomes and modelling multidimensional phenotypes, as well as creating a new population-based computational method for haplotyping. The group reached interesting results in these research fields, especially from a computational point of view.
Inductive Queries for Mining Patterns and Models (IQ)
Period: 9/2005 – 8/2008
Researchers: Aristides Gionis, Heikki Mannila, Taneli Mielikäinen, Pauli Miettinen, Panayiotis Tsaparas
Funding: EU
One of the main challenges of data mining is developing a common theoretical framework. Inductive databases, i.e. databases for data mining with a declarative approach to data analysis, offer a promising approach to this problem.
The IQ project studies the theoretical basis of inductive databases and develops inductive databases for various data analysis problems. They are being applied to biological problems.
Researchers from Belgium (Universiteit Antwerpen), France (Institut National des Sciences Appliquées de Lyon), Germany (Albert-Ludwigs-Universität Freiburg), Slovenia (Institute Josef Stefan) and Finland (Helsinki Institute for Information Technology) participate in this project.
Clustering of phenotypic features in schizophrenia and bipolar disorder samples (PhenoClusters)
Period: 8/2004 - 9/2005
Researchers: Heikki Mannila, Jaana Wessman, Mikko Koivisto, Laura Ruotsalainen
Funding: Orion Oyj
It is important to establish the genetic basis for multifactorial diseases. It is challenging to analyse the data when there is a large amount of phenotypes connected with a disease. This project develops probability models and clustering programs for multidimensional phenotype data. The methods are applied to phenotype and genotype data related to schizophrenia and bipolar disorder. The project is a joint effort with the National Public Health Institute.
In 2005, the project selected a number of the most interesting clusterings and studied their features in detail in cooperation with experts from the NPHI. The project was declared a success: the clustering approach opened up new viewpoints on multidimensional data, and exposed some formerly unknown features that will require further investigation.