The neuroinformatics group
Neuroinformatics is a combination of computer science and neuroscience, and the interface between them. There are two sub-fields in neuroinformatics. The goal of computational neuroscience is to build simulation models of brain functions; in practice, each model only represents an isolated, smaller part of the brain. The second sub-field is neuroscientific data analysis. Modern devices in e.g. brain imaging create huge amounts of data, and new methods are required for analysing them.
When it comes to methods, the research group has concentrated on probabilistic and statistical models, and in the field of neuroscience, the emphasis lies on modelling the vision system of the brain and on analysis of brain-imaging data. In addition, the group develops general methods of data analysis, which can be applied in other fields, such as bioinformatics.
The research is based on theories of non-Gaussian and unsupervised learning. In unsupervised learning, we learn new features from data that has not been classified or divided into input and output. Classical statistical methods usually assume that data is Gaussian, i.e. it follows normal distribution. If this is not true, we can develop new kinds of methods for analysis. The analysis of independent components developed in the 1990s is one of the most important examples of such methods.
Lately, we have found applications for this framework in e.g. causal modelling, where we try to model the interaction between different variables. These methods have major applications in the analysis of brain imaging data, as well as in other fields.
We are using the same theoretical framework for modelling vision systems. The premise is that the brain makes a statistical analysis of incoming information, and this analysis can be simulated on the basis of statistical models of digital images.
Contact person : Professor Aapo Hyvärinen
Homepage : www.hiit.fi/neuroinf
Selected publications
A. Hyvärinen and P.O. Hoyer. Emergence of phase and shift invariant features by decomposition of natural images into independent feature subspaces. Neural Computation, 12(7):1705-1720, 2000.
A. Hyvärinen, J. Karhunen, and E.Oja. Independent Component Analysis. Wiley Interscience, 2001. Japanese translation by Tokyo Denki University Press, 2005. Chinese translation by Electronic Industry Press, Beijing , 2007.
A. Hyvärinen. Estimation of non-normalized statistical models using score matching. Journal of Machine Learning Research, 6:695--709, 2005.
S. Shimizu , P.O. Hoyer, A. Hyvärinen, and A. Kerminen. A linear nongaussian acyclic model for causal discovery. J. of Machine Learning Research 7:2003-2030, 2006.