Maintaining population diversity by minimizing mutual information
Yong Liu, Xin Yao · 2002
Based on negative correlation learning [1] and evolutionary learning, evolutionary en-sembles with negative correlation learning (EENCL) was proposed for learning and de-signing of neural network ensembles [2]. The idea of EENCL is to regard the population of neural networks as an ensemble, and the evo-lutionary process as the design of neural net-work ensembles. EENCL used a tness shar-ing based on the covering set. Such tness sharing did not make accurate measurement on the similarity in the population. In this paper, a tness sharing scheme based on mu-tual information is introduced in EENCL to evolve a diverse and cooperative population. The eectiveness of such evolutionary learn-ing approach was tested on two real-world problems. 1