Activation Functions in Neural Networks

When we pass any function to hidden layer, known as activation function.The task of activation function is to filter, normalize and non linearize the dataset and it also fires the input to perceptron of the next layer. In neural networks we update weights and biases in reference with error at output using Back propagation. Back Propagation would be possible only because of activation function. When we will be approaching deep learning problems most of the times we would deal with complex datasets. Non linear function gives clear understanding about complex datasets and can be differentiated multiple times, will give weights and baises for different- different layers and hence, will provide better learning opportunity for neural networks.