(aside image)

My research topics

Prof. Juha Karhunen has belonged to five research groups in the new Department of Computer Science, listed and discussed briefly below. He is retiring now and the researchers under his formal supervision have been moved to other professors. From the web pages of these research groups, especially on their subpages "Research", you can find more detailed information on their research topics and results.

Applications of Machine Learning (AML) research group

In September 2013, I became the head and supervising professor of the AML research group, following Prof. Olli Simula after his retirement. The name of this research group was until recently Environmental and Industrial Machine Learning (EIML). This has been my most important research group. My tasks have involved acquiring financial support for the research group, guiding and handling doctoral and Master's (Diploma Engineer) theses, making formal decisions, arranging group meetings, and some other matters. Dr. Amaury Lendasse created and led in practice the research of the EIML group, whose name was originally time series processing (TSP) group. From September 1, 2014, Dr. Lendasse became associate professor in the Iowa University in the United States.

On the theoretical side, the AML group has been quite active in improving extreme learning machine (ELM) in various ways. In its basic form extreme learning machine, introduced in 2006, is a fast but efficient neural network method which has recently become quite popular, and has already many practical applications. The AML research group has also made theoretical contributions to time series prediction, variable selection, and projecting data into two dimensions for information visualization. The application topics of the AML research group include or have included environmental modeling, various industrial applications of machine learning, and malfare detection in internet and mobile phones. Recently, researcher and graduate student Luiza Sayfullina was moved under Assistant Prof. Alexander Jung's supervision, and similarly Alexander Grigorievskiy under Prof. Aki Vehtari's supervision. Alexander Grigorievskiy has enough papers for his doctoral thesis, and he is writing its introductory and summarizing part.

Deep Learning and Bayesian Modeling

Currently, this research group is studying mainly deep learning of complex and large probabilistic models. Deep learning deals with advanced methods for learning neural networks that have several hidden layers. It is currently a hot research topic because it has provided world record performances in several benchmark problems of machine learning. Assistant professor Tapani Raiko led this research group until he moved to Apple Corp. in August 2016. Kyunghyun Cho, Jaakko Luttinen, and Mathias Berglund have achieved their Ph.D. in technology degrees, and moved then to new positions elsewhere. Several of the researchers in the Deep Learning and Bayesian Modeling research group are now working in the Curious AI company. It is developing so-called Ladder and Tagger deep neural networks that have provided excellent results. Curious AI's CEO and main innovator Dr. Harri Valpola was the principal researcher in the predecessor of this research group described below, and got his doctoral degree under my supervision in 2000.

Some years ago this was my main research group, and its name was Bayesian algorithms for latent variable models. This former research group applied variational Bayesian learning (sometimes earlier called also Bayesian ensemble learning) methods to unsupervised or blind learning problems for continuous-valued data. The data is modeled using either neural networks or other graphical models. More detailed information on these research results can be found in the research reports of the Bayes group, covering the years 2010-2011, 2008-2009, 2006-2007, 2004-2005, 2002-2003, and 2000-2001.

Independent component analysis and blind source separation

The Independent component analysis (ICA) and blind source separation (BSS) research group has studied blind source separation, independent component analysis, and non-negative low-rank learning. The group was largest and most active around the year 2000, with publication of the book "Independent Component Analysis" which has become a standard reference in the field. Another major achievement of this research group was the development of the FastICA algorithm by current Prof. Aapo Hyvärinen in 1997. Due to its computational efficiency and effectiveness, FastICA is still the most widely used ICA algorithm. Prof. Erkki Oja has led this research group. After his retirement in February 2015 the group has not been active, but it is historically quite important.

Detailed information on earlier research results of the croup can be found in the research reports of the ICA and BSS group, covering the years 2010-2011, 2008-2009, 2006-2007, 2004-2005, 2002-2003, as well as theoretical ICA research in 2000-2001, and applications of ICA in 2000-2001.

Computational methods and data analysis for astrophysics

Juha Karhunen was the supervising professor and formal head of the Computational methods and data analysis for astrophysics research group until February 2017, until Prof. Aki Vehtari took my place. In practice, the group is led by Dr. Maarit Käpylä, earlier Mantere. She is is now 20% of her time visiting professor here at Aalto University Dept. of Computer Science. Her main affiliation is however research group leader in the highly esteemed Max Planck Institute of Solar System in Göttingen, Germany. Dr. Frederick Gent, Dr. Matthias Rheinhardt, and M.Sc. Nigul Olspert are working in her research group in our department in Aalto University, in collaboration with several researchers fron University of Helsinki. This research group belongs to the ReSoLVE (Research on Solar Long-term Variability and Effects) Centre of Excellence of the Academy of Finland.

Content-based image and information retrieval

I was also the supervising professor and formal head of the Content-based image and information retrieval research group after the retirement of Prof. Erkki Oja in February 2015. The research of this group is led in practice by Dr. Jorma Laaksonen. Its new supervising professor is Samuel Kaski.

My other and older research topics

I have studied also robust principal component analysis (PCA) and missing values in PCA, see my recent publications.

Information about my older research efforts on nonlinear PCA, ICA and BSS, and their extensions, as well as on subspace methods in CDMA can be found in my publications as well as in the triennial report (covering the years 1997-1999) and quinquennial report (covering the years 1994-1998) of our laboratory, available here.