Presented by Erkki Somersalo Professor, Department of Mathematics, Applied Mathematics, and Statistics.Case Western Reserve University
Abstract: Self-organizing map (SOM) is a classic example of early artificial neural network algorithms for dimension reduction and organizing, visualizing and analyzing data. The algorithm was proposed and developed by Teuvo Kohonen in the 1980s, and is sometimes referred to as Kohonen map. The algorithm is heuristic, drawing ideas from organization of neurons to perform special tasks, and, e.g., Hebbian learning models. It also constitutes a basis for certain classification algorithms based of feature vectors such as learning vector quantifier (LVQ). In this talk, the basic idea of SOM is explained, with the emphasis on the intuitive side of the approach related to the geometric vs. topological organization of data, and the algorithm is then applied to certain high dimensional imaging data, in particular, the hyperspectral imaging in remote sensing as well as texture analysis.