 |
    |
Evolving Femtocell Algorithms with Dynamic and Stationary Training Scenarios
Erik Hemberg1, Lester Ho2, Michael O’Neill1, and Holger Claussen2
1Natural Computing Research & Applications Group Complex & Adaptive Systems Laboratory School of Computer Science & Informatics, University College Dublin, Ireland
erik.hemberg@ucd.ie
m.oneill@ucd.ie
2Bell Laboratories, Alcatel-Lucent, Dublin, Ireland
lester.ho@alcatel-lucent.com
holger.claussen@alcatel-lucent.com
Abstract. We analyse the impact of dynamic training scenarios when evolving algorithms for femtocells, which are low power, low-cost, user-deployed cellular base stations. Performance is benchmarked against an alternative stationary training strategy where all scenarios are presented to each individual in the evolving population during each fitness evaluation. In the dynamic setup, different training scenarios are gradually exposed to the population over successive generations. The results show that the solutions evolved using the stationary training scenarios have the best out-of-sample performance. Moreover, the use of a grammar which produces discrete changes to the pilot power generate better solutions on the training and out-of-sample scenarios.
LNCS 7492, p. 518 ff.
lncs@springer.com
© Springer-Verlag Berlin Heidelberg 2012