What size net gives valid generalization?
Neural Computation
Least Squares Support Vector Machine Classifiers
Neural Processing Letters
An introduction to support Vector Machines: and other kernel-based learning methods
An introduction to support Vector Machines: and other kernel-based learning methods
SMO algorithm for least-squares SVM formulations
Neural Computation
Pattern Classification (2nd Edition)
Pattern Classification (2nd Edition)
The mee principle in data classification: A perceptron-based analysis
Neural Computation
| Hi-index | 0.00 |
We investigate the capability of multilayer perceptrons using specific risk functionals attaining the minimum probability of error (optimal performance) achievable by the class of mappings implemented by the multilayer perceptron (MLP). For that purpose we have carried out a large set of experiments using different risk functionals and datasets. The experiments were rigorously controlled so that any performance difference could only be attributed to the different risk functional being used. Statistical analysis was also conducted in a careful way. From the several conclusions that can be drawn from our experimental results it is worth to emphasize that a risk functional based on a specially tuned exponentially weighted distance attained the best performance in a large variety of datasets. As to the issue of attaining the minimum probability of error we also carried out classification experiments using non-MLP classifiers that implement complex mappings and are known to provide the best results until this date. These experiments have provided evidence that at least in many cases, by using an adequate risk functional, it will be possible to reach the optimal performance.