SOCIAL: self-organizing classifier ensemble for adversarial learning
MCS'10 Proceedings of the 9th international conference on Multiple Classifier Systems
| Hi-index | 0.00 |
In a scenario of supervised classification of data, labeled training data is essential. Unfortunately, the process by which those labels are obtained is not error-free, for example due to human nature. The aim of this work is to find out what impact noise on the labels has, and we do so by artificially adding it. An algorithm for the noising procedure is described. Not only individual classifiers are studied, but also ensembles of classifiers whose answers are combined, increasing the overall performance. Also, we will answer the question if classifiers trained on soft labels are more resilient to label noise than those trained on hard labels.