L1 penalizes the absolute value of the weight and tends to make the weights zero. L2 penalizes the squared value of the weight and tends to make the weight smaller during the training. Both the regularizes assume that models with smaller weights are better.
L1 and L2 regularization
