| Autor: |
Laxmi, Sree B. R., Vijaya, M. S. |
| Předmět: |
|
| Zdroj: |
International Journal of Simulation: Systems, Science & Technology; Dec2018, Vol. 19 Issue 6, p1-8, 8p |
| Abstrakt: |
In spite of the popularity and success rates of the Deep learning algorithms in solving complex non-linear problems, it can be observed that the imbalanced dataset contributes to the misclassification rate of the models. Studies at present merely focus on the problem with imbalanced dataset. In this paper, we propose Weighted Mean Square Error (WMSE) to handle the imbalanced dataset problem while training the Deep Belief Networks. This error metrics help in reducing the dominance of the majority classes' influence on building the classification model. The measure is evaluated against imbalanced subset of benchmark datasets MNIST (Appendix-I) and CIFAR-100 (Appendix-II); and with a Tamil phoneme dataset 'Kazhangiyam' built in our earlier work and found to build better classification models for Tamil phoneme recognition problem. [ABSTRACT FROM AUTHOR] |
| Databáze: |
Complementary Index |
| Externí odkaz: |
|
