Friday, April 17, 2015

Mark van Heeswijk: Advances in Extreme Learning Machines

Mark van Heeswijk is defending his PhD thesis "Advances in Extreme Learning Machines" at Aalto University School of Science. As the opponent serves Professor Donald C. Wunsch, Missouri University of Science and Technology, USA, and as the Custos Professor Erkki Oja, Aalto University School of Science, Department of Computer Science. A large proportion of the research reported in the thesis was conducted in the research group lead by Dr. Amaury Lendasse. Landasse serves nowadays at the University of Iowa.

Extreme learning machine (ELM) is a modification of single layer feedforward neural network. An ELM is based on the idea that the hidden neurons do not need to be trained but can be randomly generated. The name and origin of the method was discussed. The opponent referred to the recent discussions on the fact that there have been related methods before the ELM such as RVFL (Random vector functional linear network). It was concluded that the Van Heeswijk has carefully referred both to the older and more recent work.

The opponent quoted a piece of news according to which the amount of data will soon grow over zetabyte due to the developments related to the internet of things. He also reminded that neural networks have been widely deployd in different areas of society. For example, in the Humane Genome techniques have been heavily applied. In the discussion, a number of topics were covered including the role of randomness, compressive sensing, sparseness, and time-accuracy tradeoff.

No comments: