A Comparative Analysis of the Most Commonly Used Activation Functions in Deep Neural Network
R. Mahima, M. Maheswari, S. Roshana, E. B. Priyanka, Neha Mohanan, N. Nandhini
Abstract
Due to their effectiveness in many applications like neural language processing, image analysis, video analysis and healthcare and their ability to resolve challenging machine learning issues and complex problems, deep neural networks have recently gained more significance. A neural network may learn to simulate nonlinear relationships between the input and output variables by introducing nonlinearity through an activation function. As a result, it can recognize more intricate patterns and make more precise predictions. There are several activation functions, and it is used on a neural network determines how well it performs. The most important activation functions, are hyperbolic tangent (Tanh), sigmoid activation function, Rectified linear unit (ReLU), Exponential linear unit (ELU), softmax and a leaky ReLU activation function. They are analyzed and compared here. The conclusion of this research may be used in the comparison and to choose the activation functions according to the need.