Publicação
Neuroevolution under unimodal error landscapes : an exploration of the semantic learning machine algorithm
| Resumo: | Neuroevolution is a field in which evolutionary algorithms are applied with the goal of evolving Artificial Neural Networks (ANNs). These evolutionary approaches can be used to evolve ANNs with fixed or dynamic topologies. This paper studies the Semantic Learning Machine (SLM) algorithm, a recently proposed neuroevolution method that searches over unimodal error landscapes in any supervised learning problem, where the error is measured as a distance to the known targets. SLM is compared with the topology-changing algorithm NeuroEvolution of Augmenting Topologies (NEAT) and with a fixed-topology neuroevolution approach. Experiments are performed on a total of 6 real-world datasets of classification and regression tasks. The results show that the best SLM variants outperform the other neuroevolution approaches in terms of generalization achieved, while also being more efficient in learning the training data. Further comparisons show that the best SLM variants also outperform the common ANN backpropagation-based approach under different topologies. A combination of the SLM with a recently proposed semantic stopping criterion also shows that it is possible to evolve competitive neural networks in a few seconds on the vast majority of the datasets considered. |
|---|---|
| Autores principais: | Jagusch, Jan-Benedikt |
| Assunto: | Semantic Learning Machine NEAT Neuroevolution |
| Ano: | 2018 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade Nova de Lisboa |
| Idioma: | inglês |
| Origem: | Repositório Institucional da UNL |
| Resumo: | Neuroevolution is a field in which evolutionary algorithms are applied with the goal of evolving Artificial Neural Networks (ANNs). These evolutionary approaches can be used to evolve ANNs with fixed or dynamic topologies. This paper studies the Semantic Learning Machine (SLM) algorithm, a recently proposed neuroevolution method that searches over unimodal error landscapes in any supervised learning problem, where the error is measured as a distance to the known targets. SLM is compared with the topology-changing algorithm NeuroEvolution of Augmenting Topologies (NEAT) and with a fixed-topology neuroevolution approach. Experiments are performed on a total of 6 real-world datasets of classification and regression tasks. The results show that the best SLM variants outperform the other neuroevolution approaches in terms of generalization achieved, while also being more efficient in learning the training data. Further comparisons show that the best SLM variants also outperform the common ANN backpropagation-based approach under different topologies. A combination of the SLM with a recently proposed semantic stopping criterion also shows that it is possible to evolve competitive neural networks in a few seconds on the vast majority of the datasets considered. |
|---|