Research

by Francesc Alías, Xavier Llorà, Lluís Formiga, Kumara Sastry, and David E. Goldberg (2006, accepted). IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP 2006). May 14-19, 2006. Also as IlliGAL TR No 2005022. Link to the PDF. Abstract Active interactive genetic algorithms (aiGAs) rely on actively optimizing synthetic fitness functions. In interactive genetic algorithms (iGAs) framework, user evaluations provide the necessary input for synthesizing a reasonably accurate surrogate fitness function that models user evaluations or, in other words, his/her decision preferences. User evaluations collected via tournament selection only provide partial-ordering relations between solutions. Active iGAs assemble a partial-ordering graph of user evaluations. In such a directed graph, any contradictory evaluation provided by the user introduces a cycle in the graph. This property is explored in this paper to measure the consistency of the evaluations provided by the user along the evolutionary process. The consistency measures are applied to a real-world problem, the weight tuning of the cost function involved in corpus-based text-to-speech synthesis. Results show the usefulness of such measures to identify inconsistent users during the evolutionary tuning process, and successfully the number of evaluations required by more than half. ...

Recently Jack Park from Iris Semantic Network left a comment asking, among other things, if DISCUS was going open source. This is an interesting question. DISCUS is currently under two patent pending processes. However, the current policies at NCSA, one of the DISCUS partners, is to push toward open source. We are currently finishing a first release of DISCUS International with support for English and Japanese. When it passes the release quality standards, some parts of DISCUS will be disseminated, may be as open source code. This is, however, an issue that needs to be ultimately discussed with the Office of Technology Management (the university office handling the DISCUS patent process). ...

Torsten Horthorn maintains a page with a list of packages for machine learning and statistical learning in R.

This is the presentation for the initial paper on the compact classifier system published in CEC 2005 (Edimburgh., Scotland, September 2005). The presentation slides here (1Mb).

by Xavier Llorà, Kumara Sastry, and David E. Goldberg (2006). Proceedings of the Congress on Evolutionary Computation, 1, 596—603. Also as IlliGAL TR No 2005019. Link to the PDF. Abstract This paper presents an analysis of how maximally general and accurate rules can be evolved in a Pittsburgh-style classifier system. In order to be able to perform such an analysis we introduce a simple bare-bones Pittsburgh-style classifier systems—the compact classifier system (CCS)—based on estimation of distribution algorithms. Using a common rule encoding schemes of Pittsburgh-style classifier systems, CCS mantains a dynamic set of probability vectors that compactly describe a rule set. The compact genetic algorithm is used to evolve each of the initially perturbated probability vectors. Results show how CCS is able to evolve in a compact, simple, and elegant manner rule sets composed by maximally general and accurate rules. The initial theoretical analysis and results also show that traditional encoding schemes used by Pittsburgh-style classifiers add an extra facet of diffiiculty. Such a bias plays a central role on the overall performance and scalability of CCS and other Pittsburgh-style systems using such encoding schemes. ...