!-- Converted with html_pointers.awk $Revision: 1.5 $ W.B.Langdon@cwi.nl --> 1
In the paper a method that moderate code growth in genetic programming is presented. The addressed problem is symbolic regression. A special mutation operator is used for the simplification of programs. If every individual program in each generation is simplified, then performance of the genetic programming system is worsened. But if simplification is applied as a mutation operator, more compact solutions of the same or better accuracy can be obtained
Genetic programming, in conjunction with advanced analytical instruments, is a novel tool for the investigation of complex biological systems at the whole-tissue level. In this study, samples from tomato fruit grown hydroponically under both high- and low-salt conditions were analysed using Fourier-transform infrared spectroscopy (FTIR), with the aim of identifying spectral and biochemical features linked to salinity in the growth environment. FTIR spectra are not amenable to direct visual analysis, so supervised machine learning was used to generate models capable of classifying the samples based on their spectral characteristics. The genetic programming (GP) method was chosen, since it has previously been shown to perform with the same accuracy as conventional data modelling methods, but in a readily-interpretable form. Examination of the GP-derived models showed that there was a small number of spectral regions that were consistently being used. In particular, the spectral region containing absorbances potentially due to a cyanide/nitrile functional group was identified as discriminatory. The explanatory power of the GP models enabled a chemical interpretation of the biochemical differences to be proposed. The combination of FTIR and GP is therefore a powerful and novel analytical tool which, in this study, improves our understanding of the biochemistry of salt tolerance in tomato plants.
This booklet contains the late-breaking papers of the Second European Workshop on Genetic Programming (EuroGP'99) held in Göteborg Sweden 26-27 May 1999. EuroGP'99 was one of the EvoNet workshops on evolutionary computing, EvoWorkshops'99. The purpose of the late-breaking papers was to provide attendees with information about research that was initiated, enhanced, improved, or completed after the original paper submission deadline in December 1998. To ensure coverage of the most up-to-date research, the deadline for submission was set only a month before the workshop. Late-breaking papers were examined for relevance and quality by the organisers of the EuroGP'99, but no formal review process took place. The 3 late-breaking papers in this booklet (which was distributed at the workshop) were presented during a poster session held on Thursday 27 May 1999 during EuroGP'99. Authors individually retain copyright (and all other rights) to their late-breaking papers. This booklet will be available as a technical report from Centrum voor Wiskunde en Informatica, Kruislaan 413, NL-1098 SJ Amsterdam http://www.cwi.nl/static/publications/reports/reports.html
The goal is to design the 2-dimensional profile of an optical lens in order to control focalplane irradiance of some laser beam. The numerical simulations of irradiance of the beam through the lens, including some technological constraints on the correlation radius of the phase of the lens, involves two FFT (fast Fourier transforms) computations, whose computational cost heavily depends upon the chosen discretization. A straightforward representation of a solution is that of a matrix of thicknesses, based on a N by N (with N a power of two) discretization of the lens. However, even though some technical simplifications allow us to reduce the size of the search space, its complexity increased quadratically with N, making physically realistic cases (e.g. N >= 256) almost untractable (more than 2000 variables). An alternative representation is brought by GP parse trees, searching in functional space: the genotype does not depend anymore on the chosen discretization. The implementation of both parametric representation (using ES algorithms) and functional approach (using standard GP) for the lens design are described. Both achieve good results compared to the sate-of-the-art methods for small to medium values of the discretization parameter N (up to 256). Moreover, preliminary comparative results are presented between the two representations, and some counter-intuitive results are discussed.