KAEiOG - Konferencja Algorytmy Ewolucyjne i Optymalizacja Globalna

XIII Krajowa Konferencja
Warszawa, 21 - 22 września 2011

Archiwum / Archives  
Sławomir Wierzchoń, Urszula Kużelewska

ImmuNet: a new technique of data exploration based on Artificial Immune Networks

Artificial Immune Systems – AIS for short – like Artificial Neural Networks emerged as an effect of observing natural mechanisms occurring inside living organisms. From an information processing standpoint, vertebrate immune system possesses many valuable features, like its ability to learn, to memorize learned patterns, and to discriminate between self and non-self molecules. These features justify rising interest in using AIS in many domains, like fault detection, computer security, data mining, robots control, etc. – consult (de Castro and Timmis, 2002) for an extensive discussion and bibliography.

Although there is lack of any common methodology of designing AIS, the immune algorithms exploit a number of features governing the behavior of natural immune system. In general, the antigen is the problem to be solved and the antibody is the generated solution. Antibodies are specialized detectors located on the surface of so-called B-cells (i.e. lymphocytes of type B). Antibodies specialize during so-called primary immune response. At the beginning of this response the antigen (problem) is recognized by poor candidate solutions, and at the end of the primary response the antigen is defeated by good candidate solutions. The process of antibodies improvement (called affinity maturation) is governed by two important mechanisms: clonal selection and apoptosis. The first of them is a Darwinian process of proliferation and differentiation during which effective lymphocytes produce theirs copies, which are subjected intensive mutation (called somatic hypermutation). Clones that strongly bind the antigen survive and again produce their own mutated clones, while the cells weakly responding to the antigens are removed. Only those cells capable of recognizing presented antigen become long-lived memory cells. The apoptosis (or programmed death of cells) is responsible for entering new types of lymphocytes and immune system stability (Jerne, 1974). In summary, the primary immune response corresponds to training phase while so-called secondary immune response – invoked when an already presented antigen is entered again to the organism – is the testing phase where the system tries to solve already known problems or problems similar to these presented during primary response.

The aim of this paper is to present an application of the artificial immune system to data clustering tasks. The system copes with numerical multidimensional data and is equipped with data visualization unit enabling observation of a cluster formation process. An immune approach to clustering task is presented in section 2 and the authors' algorithm of learning immune network and visualization is described accordingly in section 3 and 4. Section 5 then presents the results obtained from the network, which are commented in section 6.

Inne artukuły z wybranej konferencji.
[Strona główna] [Tematyka] [Ważne terminy] [Komitet Naukowy] [Harmonogram] [Goście specjalni] [Dla autorów] [Dla uczestników] [Dla recenzentów] [Lokalizacja] [Organizatorzy] [Kontakt] [Sponsorzy] [Plakat konferencji] [Galeria] [Archiwum] 

autor: Krzysztof Adamski