Name
Abstract | ||
|---|---|---|
In this paper we discuss the use of boundary methods (BMs) for distribution analysis. We view these methods as tools which can be used to extract useful information from sample distributions. We believe that the information thus extracted has utility for a number of applications, but in particular we discuss the use of BMs as a mechanism for class separability estimation and as an aid to constructing robust and efficient neural networks (NNs) to solve classification problems. In the first case, BMs can establish the utility of a data set for classification. We demonstrate experimentally that the derived ranking is consistent with alternative ranking techniques based on Bayes error (ε). Finally, BMs are used as sample selection (SS) mechanism to train NN by means of gradient algorithms. In particular, elliptic BMs (EBMs) are used to select samples so that the initial partial training set is linearly separable. In a progressive way, new samples are added to the training set solving the problem in an incremental manner. Multi-layer perceptrons (MLPs) and radial basis functions (RBFs) have been used in this work. Our results show that the probability of being trapped in a local minimum is clearly reduced when EBMs are used, making the training independent of the initial weight values. Also, the effect of the very noisy samples and outliers is reduced when the SS-EBM algorithm is employed, so we propose this method as a robust procedure to train NNs by means of a gradient learning rule. |
Year | DOI | Venue |
|---|---|---|
2000 | 10.1016/S0925-2312(00)00293-9 | Neurocomputing |
Keywords | Field | DocType |
Classification neural systems,Class separability estimation,Bayes’ error estimation,Gradient algorithms,Sample selection strategies | Linear separability,Radial basis function,Pattern recognition,Ranking,Outlier,Learning rule,Artificial intelligence,Artificial neural network,Perceptron,Machine learning,Mathematics,Bayes' theorem | Journal |
Volume | Issue | ISSN |
35 | 1-4 | 0925-2312 |
Citations | PageRank | References |
3 | 0.40 | 10 |
Authors | ||
5 | ||
Name | Order | Citations | PageRank |
|---|---|---|---|
| José-Luis Sancho | 1 | 3 | 1.08 |
| william e pierson | 2 | 3 | 0.74 |
| batu ulug | 3 | 3 | 0.40 |
| Figueiras-Vidal, A.R. | 4 | 295 | 40.59 |
| Stanley C. Ahalt | 5 | 435 | 54.14 |