Evaluation of Gradient Descent Optimization: Using Android Applications in Neural Networks

Artificial neural networks have gained prominence by being used in different aspects such as medical diagnosis and detecting malware applications. However, neural network models could have an error rate that indicates their performance. Thus, optimization algorithms can minimize the error rate by updating the neural network parameters to reach an optimal solution. This paper explores the use of permissions and underlying Linux system information features in Android platform to evaluate gradient descent optimization algorithms in neural networks. Those optimizers are evaluated by running them on a set of Android applications to find the optimum one. Furthermore, each optimizer is assessed based on its default and adjusted parameters values. This evaluation shows that the best accuracy score is 92.21% collected by Adam optimizer.