Fetching the paper…

Parameter Efficient Training of Deep Convolutional Neural Networks by Dynamic Sparse Reparameterization · Around