Author: aorriols

The landscape contest is a research competition aimed at finding out the relation between data complexity and the performance of learners. Comparing your techniques to those of other participants may contribute to enrich our understanding of the behavior of machine learning techniques and open further research lines.

The contest will take place on August 22, during the 20th International Conference on Pattern Recognition (ICPR 2010) at Istanbul, Turkey.

We encourage everyone to participate and share with us your work! For further details about dates and submission, please see this document or visit the contest webpage.

the attached PDF document or visit the contest webpage:  http://www.salle.url.edu/ICPR10Contest/.

by Albert Orriols-Puig, Ester Bernadó-Mansilla, David E. Goldberg, Kumara Sastry, and Pier Luca Lanzi. IEEE Transactions on Evolutionary Computation, doi=10.1109/ TEVC.2009.2019829, [Publisher site].

Michigan-style learning classifier systems (LCSs) are online machine learning techniques that incrementally evolve distributed subsolutions which individually solve a portion of the problem space. As in many machine learning systems, extracting accurate models from problems with class imbalances—that is, problems in which one of the classes is poorly represented with respect to the other classes—has been identified as a key challenge to LCSs. Empirical studies have shown that Michiganstyle LCSs fail to provide accurate subsolutions that represent the minority class in domains with moderate and large disproportion of examples per class; however, the causes of this failure have not been analyzed in detail. Therefore, the aim of this paper is to carefully examine the effect of class imbalances on different LCS components. The analysis focuses on XCS, which is the most-relevant Michigan-style LCS, although the models could be easily adapted to other LCSs. Design decomposition is used to identify five elements that are crucial to guaranteeing the success of LCSs in domains with class imbalances, and facetwise models that explain these different elements for XCS are developed. All theoretical models are validated with artificial problems. The integration of all these models enables us to identify the sweet spot where XCS is able to scalably and efficiently evolve accurate models of rare classes; furthermore, facetwise analysis is used as a tool for designing a set of configuration guidelines that have to be followed to ensure convergence. When properly configured, XCS is shown to be able to solve highly unbalanced problems that previously eluded solution.

Over the last few years, the increasing interest in machine learning has resulted in the design and development of several competitive learners. Usually, the performance of these methods is evaluated by comparing the new techniques to state-of-the-art methods over a collection of real-world problems.

In early days, these comparisons followed no standard, and qualitative arguments where used to extract conclusions from the results. Although these types of analyses enabled highlighting key points about the results, they also depended, to a certain extent, on the eyes of the beholder. Therefore, the need for finding a saver framework to analyze the results arose. With these, several researchers started drawing a methodology based on statistical tests. In the last three years, the first papers appeared on that topic. One of the first contributions can be found in the paper “Statistical Comparisons of Classifiers over Multiple Data Sets” by Janez Demsar. Later on, several authors extended this first efforts to build a save environment for results analysis.

And even more recently, Francisco Herrera and his research group gathered all these efforts and made a tutorial which is available here. The tutorial explains how different tests work and draws different ways to take when applying a statistical analysis to your results.

Find below the presentation of the paper Beyond Homemade Artificial Data Sets by Núria Macià, Albert Orriols-Puig, and Ester Bernadó-Mansilla in the 2009 Hybrid Artificial Intelligence Systems (HAIS’09).

This work aims at creating boundedly difficult problems for data classification whose complexity moves through different dimensions. For this purpose, this work proposes the use of a multi-objective optimization procedure to create data sets that satisfy different criteria of complexity. Please, refer to a preprint of the paper for more information.

Tomorrow, the international Hybrid Artificial Intelligence Systems conference (HAIS) gets started in Salamanca with the special session of Knowledge Extraction based on Evolutionary Learning (KEEL). In this special session, the following 14 papers that use evolutionary algorithms for different purposes in the field of machine learning will be presented:

1. A hybrid bumble bees mating optimization – GRASP algorithm for clustering by Yannis Marinakis, Magdalene Marinaki, and Nikolaos Matsatsinis
2. A first study on the use of cooperative coevolution for instance and feature selection in classification with nearest neighbour rule by Joaquín Derrac, Salvador García, and Francisco Herrera
3. Unsupervised feature selection in high dimensional spaces and uncertainty by José R. Villar, María R. Suárez, Javier Sedano, and Felipe Mateos
4. Non-dominated multi-objective evolutionary algorithm based on fuzzy rules extraction for subgroup discovery by C. J. Carmona, P. González, M.J. del Jesus, and F. Herrera
5. A first study on the use of interval-valued fuzzy sets with genetic tuning for classification with imbalanced data-sets by J. Sanz, A. Fernández, H. Bustince, and F. Herrera
6. Feature construction and feature selection in presence of attribute interactions by Leila S. Shafti and Eduardo Pérez
7. Multiobjective evolutionary clustering approach to security vulnerability assessments by Guiomar Corral, Àlvaro Garcia-Piquer, Albert Orriols-Puig, Albert Fornells, and Elisabet Golobardes
8. Beyond homemade artificial data sets by Nuria Macià, Albert Orriols-Puig, and Ester Bernadó-Mansilla
9. A three-objective evolutionary approach to generate Mamdani fuzzy rule-based systems by Michela Antonelli, Pietro Ducange, Beatrice Lazzerini, and Francesco Marcelloni
10. A new component selection algorithm based on metrics and fuzzy clustering analysis by Camelia Serban, Andreea Vescan, and Horia F. Pop
11. Multilabel classification with gene expression programming by J. L. Ávila, E. L. Gibaja, and S. Ventura
12. An evolutionary ensemble-based method for rule extraction with distributed data by Diego M. Escalante, Miguel Angel Rodriguez, and Antonio Peregrin
13. Evolutionary extraction of association rules: A preliminary study on their effectiveness by Nicolò Flugy Papè, Jesús Alcalá-Fdez, Andrea Bonarini, and Francisco Herrera
14. A minimum-risk genetic fuzzy classifier based on low quality data by Ana M. Palacios, Luciano Sánchez, and Inés Couso

We’ll have to wait until tomorrow to know more what these promising titles hide.

A few days ago, Prof. Cirac was interviewed in a Catalan TV channel about his work on quantum theory of information. Prof. Cirac explained the method based on quantum cryptography that he and his team have been developing during the last few years, which makes sure that the information can be neither intercepted nor decrypted. Actually the information is not physically transmitted, but just appears at the receiver side.

In addition to the method itself, I was surprised by the clarity with which Prof. Cirac introduced quantum physics and reviewed some of its paradoxes. In what follows, you can find a link to the video. Unfortunately, the interview is only in Catalan (interviewer) and Spanish (Prof. Cirac).