Proceedings of STeP'96. Jarmo Alander, Timo Honkela and Matti Jakobsson (eds.),
Publications of the Finnish Artificial Intelligence Society, p. 4.

Neural Networks Research in Finland

Timo Honkela
Helsinki University of Technology
Timo Honkela@hut.fi


Short Introduction

The research related to artificial neural networks in Finland is very active. The internationally recognized position is very much based on the pioneering work by Academy Professor Teuvo Kohonen in Helsinki University of Technology. Since the 1960s, Kohonen has introduced several new concepts and algorithms to neural computing. The self-organizing map (SOM) is the best-known example. The SOM is the the most popular artificial neural network algorithm in the unsupervised learning category. Over 2000 applications of the SOM have been reported in the open literature. Professor Kohonen is still very active in the field having published recently a method for the emergence of invariant features, the Adaptive-Subspace SOM (ASSOM), and efficient creation of very large SOMs for the exploratory textual data mining in the WEBSOM method.

Two recent publications illuminate rather well the Finnish neural activities. They are ``Neural Network Research in Finland'' edited by Abhay Bulsari and published by Finnish Artificial Intelligence Society in 1993, and ``Neurolaskennan mahdollisuudet'' (in Finnish) edited by Pasi Koikkalainen and published in 1994 by TEKES (Technology Development Centre of Finland).

Printed publications reveal only the top of the iceberg while the research and development in Finland is very active. One good example is the national technology programme entitled ``Adaptive and Intelligent Systems Applications''. The programme was set up by the Technology Development Centre of Finland (TEKES). Professor Erkki Oja coordinated the planning of the programme, and the programme manager is Ossi Taipale. The budget is FIM 100 million and the programme is scheduled for 1994-1998. The research projects aim, for instance, at developing methods and tools for image and signal analysis, modelling of dynamic systems, and process control and monitoring with applications in industrial and financial processes. The ten industrial projects develop applications in multiple areas, such as fault diagnosis, fault alarm handling, tuning of intelligent controllers, and empirical modelling and analysis of large quantities of information.

The number of active research organizations, companies, and individuals is high in international comparisons even without taking into account the size of the country. The largest European conference series of the field is International Conference on Artificial Neural Networks (ICANN) that was first organised in Espoo five years ago. In the opening speech of ICANN'95 held in Paris, the chair of the conference compared the French and Finnish activities. She concluded that the applicational level is much higher in Finland. In the 1970's the spirit was kept up by professor Kohonen when otherwise the worldwide research community almost solely concentrated in developing intelligent systems by means of symbolic methods. Currently, the research related to artificial neural networks in Espoo, Helsinki, Joensuu, Lappeenranta, Oulu, Tampere, Turku, and Vaasa is well present in the major international conferences.

Further information on research and development activities in Finland can be found at the WWW address http://nucleus.hut.fi/~tho/neural/finland.html. The page contains a collection of links to organisations and some individuals in Finland.