On Iterative Neural Network Pruning, Reinitialization, and the Similarity of Masks
This work addresses the problem of understanding and optimizing neural network pruning methods for researchers and practitioners, but it is incremental as it builds on existing pruning phenomena.
The study investigated how different iterative pruning techniques affect the connectivity and learning dynamics of deep learning models, finding that magnitude-based unstructured pruning with weight rewinding induces structured-like patterns and can automatically achieve weight stability.
We examine how recently documented, fundamental phenomena in deep learning models subject to pruning are affected by changes in the pruning procedure. Specifically, we analyze differences in the connectivity structure and learning dynamics of pruned models found through a set of common iterative pruning techniques, to address questions of uniqueness of trainable, high-sparsity sub-networks, and their dependence on the chosen pruning method. In convolutional layers, we document the emergence of structure induced by magnitude-based unstructured pruning in conjunction with weight rewinding that resembles the effects of structured pruning. We also show empirical evidence that weight stability can be automatically achieved through apposite pruning techniques.