|
Hello all, both RBF and MLP are proved universal approximators. I've made some tests for some regression problems using the implementations provided by Weka, with the parameters sought by cross validation and multiple runs on random permutations of the datasets - hence the given results are statistically relevant. In all cases, the RMSE of MLP is much lower than the one of the RBF. Does anyone know a paper where such a quantitative study or comparison is made? Thanks. |
I don't know about such a paper but the relative performance depends on the dataset you are using. RBF networks are known to have troubles with high dimensional data so that could be one of the reasons for your findings.