Slawomir T. Wierzchon

Intelligent Information Processing and Web Mining

Derivation of new features of observed variables has two important goals: reduction of dimensionality and de-noising. A desired property of the derived new features is their meaningful interpretation. The SCoTLASS method (Jolliffe, Trendafilov and Uddin, 2003) offers such possibility. We explore the properties of the SCoTLASS method applied to the yeast genes data investigated in (Bartkowiak et al., 2003, 2004). All the derived features have really a simple meaningful structure: each new feature is spanned by two original variables belonging to the same block.

Immune-based algorithms for dynamic optimization

The main problem with biologically inspired algorithms (like evolutionary algorithms or particle swarm optimization) when applied to dynamic optimization is to force their readiness for continuous search for new optima occurring in changing locations. Immune-based algorithm, being an instance of an algorithm that adapt by innovation seem to be a perfect candidate for continuous exploration of a search space. In this paper we describe various implementations of the immune principles and we compare these instantiations on complex environments.

Generating Optimal Repertoire of Antibody Strings in an Artificial Immune System

In this paper an idea of the artificial immune system (or AIS for brevity) is explained. Restricting to so-called binary AIS, methods for generating a repertoire of lymphocytes of minimal size are reviewed and new algorithm of low space complexity is discussed. Besides, recipes for counting so-called holes, as well as counting the total number of unrecognizable strings are given.

Standard and Genetic k-means Clustering Techniques in Image Segmentation

Clustering or data grouping is a key initial procedure in image processing. This paper deals with the application of standard and genetic k-means clustering algorithms in the area of image segmentation. In order to assess and compare both versions of k-means algorithm and its variants, appropriate procedures and software have been designed and implemented. Experimental results point that genetically optimized k-means algorithms proved their usefulness in the area of image analysis, yielding comparable and even better segmentation results.

Studying Properties of Multipopulation Heuristic Approach to Non-Stationary Optimisation Tasks

Heuristic optimisation techniques, especially evolutionary algorithms were successfully applied to non-stationary optimisation tasks. One of the most important conclusions for the evolutionary approach was a three-population architecture of the algorithm, where one population plays the role of a memory while the two others are used in the searching process. In this paper the authors’ version of the three-population architecture is applied to four different heuristic algorithms. One of the algorithms is a new iterated heuristic algorithm inspired by artificial immune system and proposed by the authors. The results of experiments with a non-stationary environment showing different properties of the algorithms are presented and some general conclusions are sketched.

Spectral clustering based on k -nearest neighbor graph

Finding clusters in data is a challenging task when the clusters differ widely in shapes, sizes, and densities. We present a novel spectral algorithm Speclus with a similarity measure based on modified mutual nearest neighbor graph. The resulting affinity matrix reflex the true structure of data. Its eigenvectors, that do not change their sign, are used for clustering data. The algorithm requires only one parameter --- a number of nearest neighbors, which can be quite easily established. Its performance on both artificial and real data sets is competitive to other solutions.

Searching for Memory in Artificial Immune System

In this paper an idea of the artificial immune system was used to design an algorithm for non-stationary function optimization. It was demonstrated that in the case of periodic function changes the algorithm constructively builds and uses immune memory. This result was contrasted with cases when no periodic changes occur. Further, an attempt towards the identification of optimal partitioning of the antibodies population into antibodies subjected clonal selection and programmed death of cells (apoptosis) has been done.

Empirical Models for the Dempster-Shafer-Theory

In spite of many useful properties, the Dempster-Shafer Theory of evidence (DST) experienced sharp criticism from many sides. The basic line of criticism is connected with the relationship between the belief function (the basic concept of DST) and frequencies [65,18]. A number of attempts to interpret belief functions in terms of probabilities have failed so far to produce a fully compatible interpretation with DST — see e.g. [34,18,14] etc. As a way out of those difficulties, in the paper we will explain our three model proposals: (1) “the marginally correct approximation”, (2) “the qualitative model”, (3) “the quantitative model”. All of them fit the framework of DST, especially the Dempster rule of combination of evidence that was the hardest point and the point of failure of previously known attempts.

Hybrid negative selection approach for anomaly detection

This paper describes a b-v model which is enhanced version of the negative selection algorithm (NSA). In contrast to formerly developed approaches, binary and real-valued detectors are simultaneously used. The reason behind developing this hybrid is our willingness to overcome the scalability problems occuring when only one type of detectors is used. High-dimensional datasets are a great challenge for NSA. But the quality of generated detectors, duration of learning stage as well as duration of classification stage need a careful treatment also. Thus, we discuss various versions of the b-v model developed to increase its efficiency. Versatility of proposed approach was intensively tested by using popular testbeds concerning domains like computers security (intruders and spam detection) and recognition of handwritten words.

Multimodal Optimization with Artificial Immune Systems

A simple and easy to implement algorithm for multimodal function optimization is proposed. It is based on clonal selection and programmed cell death mechanisms taken from natural immune system. Empirical results confirming its usability are presented, and review of other related approaches is given.