Machine Intelligence, Generalized Rough Sets and Granular Mining: Concepts, Features and Applications

Friday 25 November 2011 - Sankar K. Pal


Friday 25 November 2011 h. 15:00, room 1AD100
Prof. Sankar K. Pal (Indian Statistical Institute, India)
"Machine Intelligence, Generalized Rough Sets and Granular Mining: Concepts, Features and Applications"

Different components of machine intelligence and their characteristics are explained. The role of rough sets in uncertainty handling and granular computing is described. The significance of its integration with fuzzy sets, called rough-fuzzy computing, as a stronger paradigm for uncertainty handling, is explained. Different applications of rough granules, significance of f-granulation and certain emerging issues in their performance are stated. Generalized rough sets using the concept of fuzziness in granules as well as in sets are defined both for equivalence and tolerance relations. Different tasks such as case generation, class-dependent rough-fuzzy granulation for classification, rough-fuzzy clustering and defining entropy and various ambiguity measures for image analysis are then addressed in this regard, explaining the nature and characteristics of granules used therein. While the method of case generation with variable reduced dimension is useful for mining data sets with large dimension and size, class dependent granulation coupled with neighborhood rough sets for feature selection is efficient in modeling overlapping classes. Significance of a new measure, called ''dispersion'' of classification performance, which focuses on confused classes for higher level analysis, is explained in this regard. Superiority of rough-fuzzy clustering is illustrated for determining bio-bases (c-medoids) in encoding protein sequence for analysis. Image ambiguity measures, which take into account both the fuzziness in boundary regions, and the rough resemblance among nearby gray levels and nearby pixels, are defined for various image analysis tasks. Merits of incorporating the concept of rough granules in gray level in addition to fuzziness for computing entropy are extensively demonstrated for image segmentation problem, as an example. The talk concludes with stating the future directions of research and challenges, and the relevance to natural computing.

Rif. int. S. Crafa

Download Seminari di Informatica