Joint Pruning & Quantization for Extremely Sparse Neural Networks
This enables efficient deployment of neural networks on low-cost, low-power hardware for applications like dense prediction tasks, though it is incremental as it builds on existing pruning and quantization techniques.
The paper tackles the problem of reducing memory and hardware costs for neural networks by jointly optimizing pruning and quantization, achieving up to 99% memory reduction and 99.9% hardware cost reduction while maintaining performance on tasks like stereo depth estimation.
We investigate pruning and quantization for deep neural networks. Our goal is to achieve extremely high sparsity for quantized networks to enable implementation on low cost and low power accelerator hardware. In a practical scenario, there are particularly many applications for dense prediction tasks, hence we choose stereo depth estimation as target. We propose a two stage pruning and quantization pipeline and introduce a Taylor Score alongside a new fine-tuning mode to achieve extreme sparsity without sacrificing performance. Our evaluation does not only show that pruning and quantization should be investigated jointly, but also shows that almost 99% of memory demand can be cut while hardware costs can be reduced up to 99.9%. In addition, to compare with other works, we demonstrate that our pruning stage alone beats the state-of-the-art when applied to ResNet on CIFAR10 and ImageNet.