Unsupervised Learning on Traditional and Distributed Systems

Giuseppe Patanè 

Objective of the thesis

The aim of this thesis is the presentation of the work and the results related to the study and the research performed during the triennium of my Ph.D. 

In this context, several new algorithms for Cluster Analysis (CA, or clustering) and Vector Quantization (VQ) [MyIWANN99ELBG], [MyWILF99ELBG], [MyNNELBG], [MyKES00], [MyFACS], [MyIJCNN00], [MyParClustering], [MyPAUL], developed during this period, will be shown.

In some cases, the techniques considered have to deal with the elaboration of large amounts of data. In such circumstances, in order to increase the available computing power, the utilization of techniques of parallel computing can be an effective solution. For this reason, with the cooperation of Prof. Marco Russo of the University of Messina (Italy), we have also worked in the construction of the MULTISOFT Machine, a powerful farm of Personal Computers employed for the implementation of several algorithms. The equipments constituting the MULTISOFT Machine have been mainly funded by the Istituto Nazionale di Fisica Nucleare (INFN) - section of Catania and also by Istituto Nazionale di Fisica della Materia (INFM) - section of Messina and Centro di Calcolo Elettronico - Università di Messina (CECUM).

In this introductory chapter, a brief description of the main themes faced will be given. It begins with the definition of the objectives of CA and VQ and the identification of several field of application. Afterwards, the new techniques developed and presented in the thesis are classified inside one of three families of algorithms for CA/VQ (traditional, incremental and parallel). Such techniques are ELBG (belonging to the family of traditional algorithms), FACS (among the incremental algorithms), PARLBG, PARELBG and PAUL (in the family of parallel algorithms). Lastly, the organization of the following chapters is described. 

Keywords: Unsupervised Learning, Vector Quantization, Clustering, Parallel, Multicomputers.

[MyPhDThesis] G.Patanè, Unsupervised Learning on Traditional and Distributed Systems, Ph.D. Thesis (in English), Palermo, Italy, 2001.

Full paper