Litcius/Paper detail

Activation Function Conundrums in the Modern Machine Learning Paradigm

Jamshaid Iqbal Janjua, Sidra Zulfiqar, Tahir Abbas Khan, Sadaqat Ali Ramay

202323 citationsDOI

Abstract

The decision-making components of main neural networks are activation functions. Furthermore, they evaluate the output of the network’s neural node; consequently, they are critical to the overall network’s performance. As a result, selecting the most appropriate activation function in neural network calculation is crucial. These functions have various features that are thought to be necessary for successful learning. These features include their methods to ensure individual derivatives and finite range. The generally used cumulative functions, such as Sigmoid, Leaky ReLU, softmax, TanH and ReLU, and so on, will be evaluated in this study article. This will be followed by their qualities, individual disadvantages and advantages, and specific formula application advice.

Topics & Concepts

Computer scienceFunction (biology)Artificial intelligenceCognitive sciencePsychologyBiologyEvolutionary biologyNeural Networks and ApplicationsMachine Learning and Data ClassificationMachine Learning and ELM