Deep learning is now the central research topic of the group. Deep learning
is a probabilistic latent variable modeling method for learning neural
networks having several layers by using layer-wise unsupervised pretraining.
Formerly we have studied variational Bayesian learning and applied it to various latent variable model structures which are suited to unsupervised learning. (See glossary for explanation of terms). We were mainly interested in models with continuous variables. Research on spatiotemporal Bayesian latent Gaussian models is still continuing.
On our research page, you can find descriptions of our current and former research topics. The Bayes chapter of the biennial report of our Adaptive Informatics Research Centre covering the years 2010-2011 contains extended summaries on our earlier research results with references mainly on variational Bayesian learning.
On our publication page, you can find selected publications of our research group downloadable in electronic form. More publications are available on their authors' homepages.
Here you can find free software packages prepared by
our research group. They include Bayes blocks and their extensions, programmed
in C++ and Python, as well as Matlab packages for nonlinear factor analysis
and nonlinear dynamical factor analysis.
Some software on deep learning algorithms developed by us are available on KyungHyun Cho's homepage
The figure on the right shows an example of the gravitational lensing phenomenon in astronomy, where several images of the source are observed. We have applied our Bayesian methods to the delay estimation problem appearing in this application. See "Applications to astronomy" on our "Research" subspage for a more detailed explanation.