Tag
This paper introduces a layerwise decoupling approach for structured sparsification of fully connected layers in neural networks, demonstrating improved robustness and reduced over-pruning compared to joint methods.