Regularization is the method to overfitting of the model. Sometimes the weights are very high this causes a big difference in the cost function with a small increment in the input values. L2 regularization heavily penalizes the high cost. The number of significant weights should be fewer. L1 regularization eliminates the insignificant weights.