Computational intelligence has been in constant development over the last decades. Every time we think that the limit has been reached, new challenges emerge. Currently, we face completely different problems than we faced 10 or 20 years ago. Dynamic streams of data, massive amounts of information being processed, real-time analyse, and social networks are among just a few of the examples that come to mind. There is a need for ongoing progress and adaptation to new problems at hand. Assumptions that general principles of given algorithms will behave similarly for each problem are not valid. The theoretical and practical sides of computational intelligence are merging. We need to have a strong background for each method and an outlook on its general performance. A high variety characterizes problems that need to be solved. It causes the need to tune each method to the specific problem. This is the most fascinating part of developing new methods—seeing them used in real-life and facing difficulties that could not be identified on the purely theoretical level.
This Special Issue includes extended versions of the best articles presented at the 7th International Symposium on Advances in Artificial Intelligence and Applications (AAIA’12), as a part of the Federated Conference on Computer Science and Information Systems 2012 (FedCSIS 2012). AAIA is a growing conference, dedicated to presenting a broad outlook on current developments in both theoretical and practical aspects of artificial intelligence. By maintaining a wide scope of topics of interest, it aims to bring together researchers working in different areas of intelligent systems and to stimulate discussion about possible intersections and hybridizations of these fields. AAIA 2012 took place in the beautiful city of Wrocław, Poland, September 9–12, 2012. From the presented articles, five were selected for presentation in this Special Issue. The choice was based on the quality and originality of the presented work and the successful blend of new theoretical findings with well-defined areas of application. Each article was significantly extended and went through a rigorous peer review, which resulted in a high–quality selection of research manuscripts. Let us describe briefly each of the articles collected herein.
The first article, by Janusz and Ślężak, deals with the problem of using rough set theory for clustering features in order to create a representation of original data in a reduced feature space. They propose a novel heuristics for calculating reduct, based on greedy search and a diverse attribute selection. Their method is presented in relation to high-dimensional datasets, in which traditional reduct generation methods display high computational complexity or fail because of the large feature space size. The main area of application presented by the authors is gene clustering and classification, an important area in bioinformatics. Their method simultaneously decreases the computational time and improves the reduct quality by finding clusters of irrelevant features.
The second article, by Sarkar, Cooley, and Srivastava, focuses on a similar problem of reducing the size of the feature space—however, from the feature-selection point of view. They introduce a robust feature-selection method and a model-selection technique, suited for high-dimensional data. The idea of the presented method is based on using inter-ranker agreement along with rank aggregation for feature selection. A scoring mechanism named the Robustness Index is used for evaluating the performance of the selectors. It measures the robustness of the algorithm based on varying training data. The presented approach is applied to the real-life problem of detecting acute myelogenous leukemia.
The third article, by Krawczyk and Woźniak, discusses the problem of forming efficient ensembles of one-class classifiers. One-class classification is a specific machine learning area, in which, during the training phase, only objects from a single class are available. Most one-class classifiers base their decision on the distance between the new object and an estimated decision boundary. To perform a classifier fusion, one needs to apply a mapping from distance to probability in order to produce support-function values. The authors show that the used distance measure has a crucial impact on the quality of the ensemble and that commonly used Euclidean distance is not always the best solution.
The fourth article, by Adany and Tamir, concentrates on introducing a novel online algorithm for battery utilization in electric vehicles. They consider the problem of utilizing a pack of batteries serving current demands in a given number of electric vehicles. The authors show that the offline problem, in which the sequence of current demands is known in advance, is strongly NP-hard. Then, they formulate an online version of this task and present a competitive algorithm associated with redundant penalty and its lower error bound.
The final article, by Anghinolfi and colleagues, presents an application of an ant colony optimization procedure for the problem of robot skin wiring. This study focuses on addressing the problem of designing a good pheromone structure. The authors propose five alternatives, which are thoroughly validated, using problem instances originating from both real and theoretical (i.e., artificially generated) use cases.
We express our gratitude to Professor Robert Trappl, Editor-in-Chief of Applied Artificial Intelligence, for allowing this Special Issue to appear in his esteemed journal. We would like to thank the reviewers for their time and most valuable help, which led to a significant improvement in the quality of the articles. We thank the authors for their interesting presentations during AAIA′12 and their quality contributions to this issue. Finally, we would like to thank Marcin Paprzycki and Maria Ganzha, chairs of FedCSIS, for their continuous support of AAIA.