SparseOpt: Addressing Normalization-induced Gradient Skew in Sparse Training

Mohammed Adnan, Rohan Jain, Tom Jacobs, Ekansh Sharma, Rahul G Krishnan, Rebekka Burkholz, Yani Ioannou

Keywords: sparse training, dynamic sparse training, training dynamics, normalization layers

TL;DR: We show that Batch Normalization has an adverse affect on the sparse training dynamics; to address this we propose a new sparsity aware optimizer to improve the sparse training dynamics and convergence rates.

Abstract: Dynamic Sparse Training (DST) methods train neural networks by maintaining sparsity while dynamically adapting the network topology. Despite the promise of reduced computation, DST methods converge significantly slower than dense training, often requiring comparable training time to achieve similar accuracy. We demonstrate both analytically and empirically that Batch Normalization (BN) adversely affects sparse training and propose SparseOpt — a sparsity-aware optimizer— to address this. Experiments on ResNet models across CIFAR-100 and ImageNet demonstrate consistently faster convergence and improved generalization with our proposed method. Our work highlights the limitations of current normalization layers in sparse training and provides the first systematic study of the interaction between Batch Normalization, sparse layers, and DST, taking a significant step toward making DST practically competitive with dense training.

Submission Number: 34