Reunión de trabajo de Geneura, del 22 de febrero de 2019

JJ nos expone el trabajo “A Neural Network System for Transformation of Regional Cuisine Style” en nuestra reunión semanal.

reunion-geneura-22-febrero-2019

Anuncios

Monitorización y clasificación de estancias en edificios empleando captación de comunicaciones inalámbricas de dispositivos inteligentes.

Dentro del CAEPIA 2018 está teniendo lugar el congreso MAEB 2018 , donde se ha presentado el trabajo Monitorización y clasificación de estancias en edificios empleando captación de comunicaciones inalámbricas de dispositivos inteligentes enmarcado por el Proyecto MOMOFES financiado por la Dirección General de Tráfico.

La presentación se encuentra a continuación:

 

Stateless evolutionary algorithms

Most algorithms keep some kind of state: global variable that holds the optimum, a counter of the number of evaluations, some context every piece algorithm must be aware of. However, this might not be the best when we want to create cloud-native algorithms, and it’s not in the case of cloudy evolutionary algorithms. There was a bit of that in GECCO, but as long as I was attending the Perl Conference in Glasgow, and I was using Perl, I kind of switched focus from the evolutionary part (but there was a bit of that too) to the language-design part and talked about evolutionary algorithms in Perl 6. The presentation is linked from the talk description.
Main problem is that you have to create dataflows that allow the algorithm to progress, as well as work efficiently in that kind of concurrent architecture, which is similar to the serverless architecture that is our eventual target.
We’ll be continuing this research in the workshop on engineering applications in Medellín, where my keynote will deal with this same topic.

Torres de la universidad//embedr.flickr.com/assets/client-code.js

Improved Genetic Fuzzy Drivers presented at CIG 2018

Last week I presented at IEEE CIG 2018 (held in Maastricht, The Netherlands) our following step in our research about autonomous drivers for Car Racing Simulators, such as TORCS, titled “The Evolutionary Race: Improving the Process of Evaluating Car Controllers in Racing Simulators“.

As commented before by @jjmerelo and later by @fergunet, we designed with Mohammed Salem (University of Mascara) a driver’s AI in which two Fuzzy Subcontrollers were hybridized with a Genetic Algorithm.

In this work we present a better evaluation approach for the GA, combining three methods: heuristic track choosing, improved fitness functions, and race-based selection of the best.

The abstract of the work is:

Simulated car races have been used for a long time as an environment where car controlling algorithms can be tested; they are an interesting testbed for all kinds of algorithms, including metaheuristics such as evolutionary algorithms. However, the challenge in the evolutionary algorithms is to design a reliable and effective evaluation process for the individuals that eventually translates into good solutions to the car racing problem: finding a controller that is able to win in a wide range of tracks and with a good quantity of opponents. Evaluating individual car controllers involves not only the design of a proper fitness function representing how good the car controller would be in a competitive race, but also the selection of the best solution for the optimization problem being solved; this decision might not be easy when uncertainty is present in the problem environment; in this case, weather and track conditions as well as unpredictable behavior of other drivers. Creating a methodology for the automatic design of the controller of an autonomous driver for a car racing simulator such as TORCS is an optimization problem which offers all these challenges. Thus, in this paper we describe an analysis and some proposals to improve the evaluation of optimized fuzzy drivers for TORCS over previous attempts to do so. It builds on preliminary results obtained in previous papers as a baseline and aims to obtain a more competitive autonomous driver via redesign of the fitness evaluation procedure; to this end, two different fitness functions are studied in several experiments, along with a novel race-based approach for the selection of the best individual in the evolution.

And the presentation is:

You can check our paper in the proceedings of the conference.

Enjoy it!

(And cite us as usual :D)

GECCO posters: modern evolutionary algorithms and particle swarm optimization methodologies

Besides the two papers we presented in GECCO workshops, our research group also had a couple of posters in the main track. Posters get a two-page publication that you can find if you want, but probably the posters themselves will be much more informative.
The first one, with Mario García, presented a new (almost) serverless architecture for evolutionary algorithms:

The second paper, with Juanlu García, Carlos Fernandes, present a structured population approach to avoid premature convergence problems with Particle Swarm Optimization algorithms

This last work shows that using a regular population structure is better for low degree of connectivity, but this degree is quite important and has a big influence on the results.
Ready to serve customers
As usual, customers received beautiful origami after listening to the explanation. Visit us next time!

A data-mining based process to early identify breast cancer from metabolomic data

Abstract of our work presented at EURO 2018, the largest and most important conference for Operational Research, co-authored by Víctor M. Rivas Santos, jointly with researchers of Complejo Hospitalario de Jaén and Fundación Medina.

This paper was presented last 9-July-2018 at Valencia, as part of the stream Data Mining and Statistics.

A data-mining based process to early identify breast cancer from metabolomic data

Abstract

We present the results yielded by our multidisciplinary group in the task of discriminating blood samples coming from breast cancer patients and healthy people. Models used to classify samples have been built using data mining techniques; data have been collected by means of liquid chromatography-mass spectrometry, a technique that detects and quantifies the metabolites present in blood samples.

Different algorithms have been tested under 10-CV and 75/25 scenarios. Our experiments showed that IBk, and J48 and Logistic Model Trees yielded rates greater than 90% only for healthy people. Naive Bayes and Random Forest enhanced the previous results in the 10-CV approach, but they did not yield more than 85% of true positives for patients in the 75/25 one. Finally, Bayesian network resulted to be the best algorithm as rates greater than 90% were yielded for both patients and rest of the people.

Many statistics have been computed as well as confusion matrices, showing that the model built by Bayesian network can effectively be used to solve this problem. Currently, the metabolites used to do built the model are being identified by biochemists. This last step will be definitive in order to consider them as a valid biomarker for breast cancer.