4th Nordic Summer School
in Geographic Information Science
Contemporary research on urban and regional systems is blessed with a profusion of data on a great variety of semantic dimensions, at a fine granularity, and often with temporal scope. Reducing the complexity of large scale datasets and uncovering salient structures out of the data noise are tasks that contribute to knowledge creation along a line of scientific reasoning that departs from earlier traditions. A variety of methodologies of knowledge discovery can enhance spatial science research. This workshop will discuss the specific contribution of Kohonen¡¯s self-organizing maps (SOM). As data reduction techniques, SOMs have been used for some time in remote sensing. It is only recently however that social scientists have become aware of the potential of self-organizing maps in understanding space-time structures that underpin urban and regional systems. The workshop will highlight the properties of self-organizing maps that make them attractive for research in social spatial sciences, in contrast to other approaches to knowledge discovery and other methods of data reduction. It will review urban and regional research that has incorporated self-organizing maps, illustrating with research conducted by the workshop instructor and others.
An important component of the workshop is that it lays the arguments according to which self-organizing maps are effective constructs for the discovery of the latent structures of the attribute space, and therefore contributes to the conceptualization and representation of ¡®space¡¯ as defined by relations between objects, rather than space as a container. This presents a departure from the Kantian view of absolute spatiality, espoused by mainstream spatial analysis research ever since Tobler¡¯s First Law of Geography stating that ¡°everything is related to everything else, but near things are more related than distant things¡± (Tobler 1970). The geographic space may not be that special after all!
Finally, the workshop discusses the implications of the notion of relative spatiality on the design of Geographic Information Systems and on spatial analysis research. Topics covered in the workshop include: 1. Principles of knowledge discovery in databases and spatial data mining 2. Spatial data visualization and visual data mining 3. Self-organizing maps: algorithm and visualization 4. SOM in urban and regional research 5. Conceptualization of space in urban & regional analysis 6. Absolute versus relative spatiality 7. Towards a paradigm shift in GIS and spatial analysis