Magnitude Attention-based Dynamic Pruning
This work addresses the inefficiency of pruning in neural networks for researchers and practitioners, offering an incremental improvement over existing methods.
The paper tackles the problem of static pruning methods that only use weight importance for structure search, by proposing a dynamic pruning approach that applies weight importance throughout training to explore sparse structures and then exploit them, achieving performance comparable to dense models and outperforming previous methods on CIFAR-10/100 and ImageNet.
Existing pruning methods utilize the importance of each weight based on specified criteria only when searching for a sparse structure but do not utilize it during training. In this work, we propose a novel approach - \textbf{M}agnitude \textbf{A}ttention-based Dynamic \textbf{P}runing (MAP) method, which applies the importance of weights throughout both the forward and backward paths to explore sparse model structures dynamically. Magnitude attention is defined based on the magnitude of weights as continuous real-valued numbers enabling a seamless transition from a redundant to an effective sparse network by promoting efficient exploration. Additionally, the attention mechanism ensures more effective updates for important layers within the sparse network. In later stages of training, our approach shifts from exploration to exploitation, exclusively updating the sparse model composed of crucial weights based on the explored structure, resulting in pruned models that not only achieve performance comparable to dense models but also outperform previous pruning methods on CIFAR-10/100 and ImageNet.