Efficient Backpropagation with Variance-Controlled Adaptive Sampling
This work addresses the computational bottleneck in training large neural networks, offering a method to speed up backpropagation with controlled accuracy, though it is incremental as it builds on existing sampling-based approaches.
The paper tackles the problem of accelerating neural network training by reducing computational costs during backpropagation, achieving up to 73.87% FLOPs reduction in backpropagation and 49.58% overall training FLOPs reduction while preserving accuracy across vision and natural language tasks.
Sampling-based algorithms, which eliminate ''unimportant'' computations during forward and/or back propagation (BP), offer potential solutions to accelerate neural network training. However, since sampling introduces approximations to training, such algorithms may not consistently maintain accuracy across various tasks. In this work, we introduce a variance-controlled adaptive sampling (VCAS) method designed to accelerate BP. VCAS computes an unbiased stochastic gradient with fine-grained layerwise importance sampling in data dimension for activation gradient calculation and leverage score sampling in token dimension for weight gradient calculation. To preserve accuracy, we control the additional variance by learning the sample ratio jointly with model parameters during training. We assessed VCAS on multiple fine-tuning and pre-training tasks in both vision and natural language domains. On all the tasks, VCAS can preserve the original training loss trajectory and validation accuracy with an up to 73.87% FLOPs reduction of BP and 49.58% FLOPs reduction of the whole training process. The implementation is available at https://github.com/thu-ml/VCAS .