Publicação
Ensembles of artificial neural networks with heterogeneous topologies
| Resumo: | Within the Machine Learning field, the emergence of ensembles, combinations of learning models, has been boosting the performance of several algorithms. Under this context, Artificial Neural Networks (ANNs) make a fruitful arena, once they are inherently stochastic. In this work, ensembles of ANNs are approached, being used several output combination methods and two heuristic ensemble construction strategies. These were applied to real world classification and regression tasks. The results reveal some improvements of ensembles over single ANNs, favoring the combination of ANNs with distinct complexity (topologies) and the weighted averaging of the outputs as the combination method. The proposed approach is also able to perform automatic model selection. |
|---|---|
| Autores principais: | Rocha, Miguel |
| Outros Autores: | Cortez, Paulo; Neves, José |
| Assunto: | Ensembles Multilayer Perceptrons Classification Regression |
| Ano: | 2004 |
| País: | Portugal |
| Tipo de documento: | comunicação em conferência |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade do Minho |
| Idioma: | inglês |
| Origem: | RepositóriUM - Universidade do Minho |
| Resumo: | Within the Machine Learning field, the emergence of ensembles, combinations of learning models, has been boosting the performance of several algorithms. Under this context, Artificial Neural Networks (ANNs) make a fruitful arena, once they are inherently stochastic. In this work, ensembles of ANNs are approached, being used several output combination methods and two heuristic ensemble construction strategies. These were applied to real world classification and regression tasks. The results reveal some improvements of ensembles over single ANNs, favoring the combination of ANNs with distinct complexity (topologies) and the weighted averaging of the outputs as the combination method. The proposed approach is also able to perform automatic model selection. |
|---|