Automatic Network Adaptation for Ultra-Low Uniform-Precision Quantization
This work addresses the challenge of maintaining inference accuracy for hardware-efficient quantization in deep learning, offering a domain-specific improvement for neural network deployment.
The paper tackles the problem of accuracy degradation in ultra-low uniform-precision neural network quantization by proposing a neural architecture search method that selectively expands channels in sensitive layers, achieving state-of-the-art 2-bit quantization accuracy on CIFAR10 and ImageNet with reduced FLOPs and parameters.
Uniform-precision neural network quantization has gained popularity since it simplifies densely packed arithmetic unit for high computing capability. However, it ignores heterogeneous sensitivity to the impact of quantization errors across the layers, resulting in sub-optimal inference accuracy. This work proposes a novel neural architecture search called neural channel expansion that adjusts the network structure to alleviate accuracy degradation from ultra-low uniform-precision quantization. The proposed method selectively expands channels for the quantization sensitive layers while satisfying hardware constraints (e.g., FLOPs, PARAMs). Based on in-depth analysis and experiments, we demonstrate that the proposed method can adapt several popular networks channels to achieve superior 2-bit quantization accuracy on CIFAR10 and ImageNet. In particular, we achieve the best-to-date Top-1/Top-5 accuracy for 2-bit ResNet50 with smaller FLOPs and the parameter size.