It is an optimisation technique used to minimise logistic regression cost function by iteratively finding the gradient of the function with the model parameters (weights and bias) and updating the parameters accordingly.