Propagation of Knowledge from Crisp and Soft Clustering through a Granular Hierarchy
Pawan Lingras
Saint Mary’s University, Halifax, Canada 

 [Abstract] Information granules is a flexible instrument for representing an object. By controlling level of abstraction, one can represent a variety of objects in a granular environment. These information granules can also be linked to each other forming granular graphs. For example, in a retail company one can represent stores, customers, customer visits, and products as information granules. A customer could be represented by an information granule consisting of data about spending, loyalty, and products. Similarly, products can be represented by the revenues, profits, and popularity. A customer granule could be connected to all the product granules based on purchases made. Analysis of data mining results in such a granular graph can provide a different perspective on the knowledge discovery. In this paper, we focus on a special case of granular graphs - hierarchical trees. In a granular hierarchy, information can be granulated at different refinement levels. For example, in a mobile phone network, a phone user is a coarser information granule compared to phone calls. A phone user granule consists of a number of phone call granules. These granular relationships can be represented using a granular hierarchy. If we apply data mining technique such as clustering to the phone users, we can not only get the profiles of the phone users from the resulting clustering, but we can propagate the clustering schemes through the granular hierarchy to obtain profiles of corresponding phone calls made by those phone users. Similarly, the knowledge obtained from the clustering schemes of finer granules such as phone numbers can be propagated up the granular hierarchy to study the profiles of corresponding phone users. This study describes transformation of crisp and fuzzy clustering schemes between different levels of granularity using mobile phone calls as a case study. The mobile phone call dataset is viewed at two different levels of granularity: finer granules corresponding to the phone calls which can be aggregated into coarser granules given by the phone numbers. Correlating results from different levels of granularity can improve the quality of analysis. The talk will describe transferring crisp clustering schemes from coarser granularity to crisp clustering at finer granularity. The difficulty in transferring crisp clustering schemes from finer granularity to crisp clustering at coarser granularity are alleviated by using fuzzy clustering. Furthermore, fuzzy clustering schemes are transferred up and down a granular hierarchy, which clearly shows that the granular transformation is more naturally applicable to fuzzy clustering than crisp clustering. The talk also shows how rough clustering can be useful in comparing fuzzy and crisp clustering schemes.

 [Biography]

Lattice Computing for Hybrid Intelligent Systems
Manuel Graña
University of the Basque Country, Spain

 [Abstract]  Lattice Theory deals with ordered sets, which are endowed with complete operations infimum and supremum. Lattice Computing is the art of algorithm design based on Lattice Theory, emphasizing order as the underlying computational framework, in contrast to conventional statistics emphasis in event frequency.  The most successful example of Lattice Computing approaches is Mathematical Morphology for image and signal processing. In the last decade Lattice Computing is extending its application to the design of intelligent systems and Machine Learning applications. Key to these advances are the definition of Morphological Auto-associative Memories endowed with surprising properties, such as perfect recall of unlimited number of patterns, and the definition of new data representations, such as the lattice of interval numbers allowing for the easy representation of multi-modal data in a common computational framework. These fundamental approaches allow easy integration into Hybrid Intelligent Systems, combined with other conventional machine learning components such as fuzzy systems or statistical or bio-inspired machine learning classifiers. Algorithms for signal transformation, feature extraction, and supervised and unsupervised classification have been defined on this Lattice Computing framework and tested on diverse application domains, such as remote sensing hyperspectral images, mobile robot navigation, medical image processing and face recognition. This presentation will summarize the main theoretical developments, and describe  a number of examples of working systems.

 [Biography]

Atomic and molecular level modeling of genetic strings using generalized RLC circuits
R. Marshall
Plymouth State University, New Hampshire, USA

 [Abstract]  We use generalized resistor-inductor-capacitor (RLC) circuits to model bases, nucleosides and nucleotides at the atomic and molecular levels and analyze the responses of these circuits to different types of input signals. Based on simple recurrence equations that characterize the impedances of such circuits, we develop compositions of these basic circuits to model arbitrarily long nucleotide strings representing DNA sequences and analyze the circuits to see if any putative connections can be established between circuit behavior and actual biological phenomena. The circuits' responses are then used to generate highly textured patterns which can be used in a variety of applications including DNA sequence mutation comparisons and assorted biometric identification schemes.

 [Biography] R. Marshall has been teaching and conducting research in Computer Science for the past 30 years. He has taught at a variety of universities including Johns Hopkins University, Boston University, Loyola University and the University of Massachusetts. He was educated at IIT-Madras (B.Tech), Dalhousie University (MS), University of Nebraska (PhD) and McGill University (ABD). He is Professor of Computer Science at Plymouth State University in New Hampshire. He has published over 120 refereed journal and international conference articles and is the author of a monograph on natural language processing and co-author of a book on distributed database systems. He has been the recipient of two Fulbright Senior Scholar awards, McConnell Fellowship, Wachovia Research Award and Hanes Sigma Delta Theta distinguished professorship. He has obtained research grants from NASA, NSF, USAID and the Department of the Navy and has held several NASA/Navy-ASEE summer research fellowships at the Naval Research Laboratory, Naval Underwater Systems Center, NASA-Goddard Space Flight Center and the Applied Physics Lab.