Role derivative of sigmoid function in neural networks

The use of derivatives in neural networks is for the training process called backpropagation. This technique uses gradient descent in order to find an optimal set of model parameters in order to minimize a loss function.