Online gradient method with smoothing ℓ0 regularization for feedforward neural networks

ℓp regularization has been a popular pruning method for neural networks. The parameter p was usually set as 0