Pruning Deep Neural Networks from a Sparsity Perspective
This work addresses the challenge of efficiently pruning deep neural networks for deployment on small devices, offering an incremental improvement with a new metric to guide pruning.
The authors tackled the problem of deep network pruning lacking a quantifiable measure for compressibility, which can lead to under- or over-pruning, by proposing the PQ Index (PQI) to estimate compressibility and developing the Sparsity-informed Adaptive Pruning (SAP) algorithm, resulting in superior compression efficiency and robustness compared to methods like lottery ticket-based pruning.
In recent years, deep network pruning has attracted significant attention in order to enable the rapid deployment of AI into small devices with computation and memory constraints. Pruning is often achieved by dropping redundant weights, neurons, or layers of a deep network while attempting to retain a comparable test performance. Many deep pruning algorithms have been proposed with impressive empirical success. However, existing approaches lack a quantifiable measure to estimate the compressibility of a sub-network during each pruning iteration and thus may under-prune or over-prune the model. In this work, we propose PQ Index (PQI) to measure the potential compressibility of deep neural networks and use this to develop a Sparsity-informed Adaptive Pruning (SAP) algorithm. Our extensive experiments corroborate the hypothesis that for a generic pruning procedure, PQI decreases first when a large model is being effectively regularized and then increases when its compressibility reaches a limit that appears to correspond to the beginning of underfitting. Subsequently, PQI decreases again when the model collapse and significant deterioration in the performance of the model start to occur. Additionally, our experiments demonstrate that the proposed adaptive pruning algorithm with proper choice of hyper-parameters is superior to the iterative pruning algorithms such as the lottery ticket-based pruning methods, in terms of both compression efficiency and robustness.