Combining Regularization and Dropout Techniques for Deep Convolutional Neural Network

Abstract

Deep learning techniques face the problem of overfitting due to their complex layer structure. Regularization methods are used to overcome this problem and improve the designed models. In this article, we use the combination of L1 regularization, L2 regularization, Elastic Net-regularization, and Dropout methods. The designed deep model using combination of these methods is considered with different rates. The deep network model using a combination of these methods is designed with different rates. Finally, the performance of all combination methods is compared with the Convolutional Neural Network model which does not use regularization methods. Experiments are performed using the Gold price per ounce data set and linear simulation model. The obtained results show that the performance of the combination model of Dropout and Elastic Net regularization is better than the other models.