L.N. de Castro

Learning and optimization using the clonal selection principle

The clonal selection principle is used to explain the basic features of an adaptive immune response to an antigenic stimulus. It establishes the idea that only those cells that recognize the antigens (Ags) are selected to proliferate. The selected cells are subject to an affinity maturation process, which improves their affinity to the selective Ags. This paper proposes a computational implementation of the clonal selection principle that explicitly takes into account the affinity maturation of the immune response. The general algorithm, named CLONALG, is derived primarily to perform machine learning and pattern recognition tasks, and then it is adapted to solve optimization problems, emphasizing multimodal and combinatorial optimization. Two versions of the algorithm are derived, their computational cost per iteration is presented, and a sensitivity analysis in relation to the user-defined parameters is given. CLONALG is also contrasted with evolutionary algorithms. Several benchmark problems are considered to evaluate the performance of CLONALG and it is also compared to a niching method for multimodal function optimization.

Artificial immune systems as a novel soft computing paradigm

AbstractArtificial immune systems (AIS) can be defined as computational systems inspired by theoretical immunology, observed immune functions, principles and mechanisms in order to solve problems. Their development and application domains follow those of soft computing paradigms such as artificial neural networks (ANN), evolutionary algorithms (EA) and fuzzy systems (FS). Despite some isolated efforts, the field of AIS still lacks an adequate framework for design, interpretation and application. This paper proposes one such framework, discusses the suitability of AIS as a novel soft computing paradigm and reviews those works from the literature that integrate AIS with other approaches, focusing ANN, EA and FS. Similarities and differences between AIS and each of the other approaches are outlined. New trends on how to create hybrids of these paradigms and what could be the benefits of this hybridization are also presented.

Data Clustering with Particle Swarms

This paper presents a new proposal for data clustering based on the particle swarm optimization (PSO) algorithm. The human tendency of adapting its behavior due to the influence of the environment minimizing the differences in opinions and ideas through time and taking into account the past experiences characterizes an emergent social behavior. In the PSO algorithm, each individual in the population searches for a solution taking into account the best individual in a certain neighborhood and its own past best solution as well. In the present work, the PSO algorithm was adapted to position prototypes (particles) in regions of the space that represent natural clusters of the input data set. The proposed method, named particle swarm clustering (PSC) algorithm, was applied in an unsupervised fashion to a number of benchmark classification problems and to one bioinformatics dataset in order to evaluate its performance.

Evolutionary algorithms for clustering gene-expression data

This work deals with the problem of automatically finding optimal partitions in bioinformatics datasets. We propose incremental improvements for a clustering genetic algorithm (CGA) culminating in the evolutionary algorithm for clustering (EAC). The CGA and its modified versions are evaluated in five gene-expression datasets, showing that the proposed EAC is a promising tool for clustering gene-expression data.

A Neuro-Immune Network for Solving the Traveling Salesman Problem

Many combinatorial optimization problems belong to the NP class and, thus, cannot be solved optimally in feasible time using standard techniques (e.g., enumeration methods). NP problems have been tackled with some success by techniques known as meta-heuristics. The present paper proposes a new meta-heuristics for solving traveling salesman problems (TSP) based on a neural network trained using ideas from the immune system. The network is self-organized and the learning algorithm aims at locating one network cell at each position of a city of the TSP instance to be solved. The pre-defined network neighborhood is going to establish the final route proposed for the TSP. The algorithm is applied to several instances from the literature and the results compared with the best solutions available.

Immune, swarm, and evolutionary algorithms. Part I: basic models

These two papers have three main aims. First (Part I), to review the general algorithms of immune, swarm and evolutionary systems. Second (Part II), to present a philosophical discussion about the similarities and differences between these paradigms, in terms of components, architecture, adaptation, interactions, and metaphors. Finally (Part II), to highlight the main features embodied in each approach, such that avenues for the creation of hybrid models can be suggested.These two papers have three main aims. First (Part I), to review the general algorithms of immune, swarm and evolutionary systems. Second (Part II), to present a philosophical discussion about the similarities and differences between these paradigms, in terms of components, architecture, adaptation, interactions, and metaphors. Finally (Part II), to highlight the main features embodied in each approach, such that avenues for the creation of hybrid models can be suggested.

An intrusion detection system using ideas from the immune system

This paper proposes an intrusion detection framework and presents a prototype for an intrusion detection system based on it. This framework takes architectural inspiration from the human immune system and brings desirable features to intrusion detection systems, such as automated intrusion recovery, attack signature extraction, and potential to improve behavior-based detection. These features are enabled through intrusion evidence detection. The prototype, called ADENOIDS, is designed to deal with application attacks, extracting signature for remote buffer overflow attacks. The framework and ADENOIDS are described and experimental results are presented.

Immune, swarm, and evolutionary algorithms. Part II: philosophical comparisons

For Part I see ibid. vol. 3 (2002). In the first part of this paper, the standard evolutionary, immune, and swarm algorithms were reviewed. This second part starts by presenting a philosophical discussion about some similarities and differences among the various approaches in terms of their basic components, structure, knowledge storage, adaptation paradigm, interactions, and metaphor. Then, the identification of the main features of each technique is performed in order to shed some light into how to create hybrid algorithms.For Part I see ibid. vol. 3 (2002). In the first part of this paper, the standard evolutionary, immune, and swarm algorithms were reviewed. This second part starts by presenting a philosophical discussion about some similarities and differences among the various approaches in terms of their basic components, structure, knowledge storage, adaptation paradigm, interactions, and metaphor. Then, the identification of the main features of each technique is performed in order to shed some light into how to create hybrid algorithms.

An evolutionary clustering technique with local search to design RBF neural network classifiers

Radial basis function neural networks constitute one type of feedforward neural net that requires a suitable determination of the basis functions so as to work properly. Among the many approaches available in the literature, the one proposed here combines a clustering genetic algorithm with K-means to automatically select the number and location of basis functions to be used in the RBF network. Preliminary simulation results suggest that the proposed hybrid algorithm can be successfully applied to classification problems, leading to parsimonious solutions, with competitive classification rates, when compared with other approaches from the RBF literature.

An Immune and a Gradient-Based Method to Train Multi-Layer Perceptron Neural Networks

Multi-layer perceptron (MLP) neural network training can be seen as a special case of function approximation, where no explicit model of the data is assumed. In its simplest form, it corresponds to finding an appropriate set of weights that minimize the network training and generalization errors. Various methods can be used to determine these weights, from standard optimization methods (e.g., gradient-based algorithms) to bio-inspired heuristics (e.g., evolutionary algorithms). Focusing on the problem of finding appropriate weight vectors for MLP networks, this paper proposes the use of an immune algorithm and a second-order gradient-based technique to train MLPs. Results are obtained for classification and function approximation tasks and the different approaches are compared in relation to the types of problems they are more suitable for.