Adaptive quantization with mixed-precision based on low-cost proxy
This work addresses the problem of deploying complex models on constrained devices, offering a practical shortcut for quantization with incremental improvements in speed.
The paper tackles efficient neural network quantization for resource-limited hardware by proposing LCPAQ, a method that uses a low-cost proxy and adaptive mixed-precision to reduce search time to 1/200 of existing methods while maintaining or improving accuracy on ImageNet.
It is critical to deploy complicated neural network models on hardware with limited resources. This paper proposes a novel model quantization method, named the Low-Cost Proxy-Based Adaptive Mixed-Precision Model Quantization (LCPAQ), which contains three key modules. The hardware-aware module is designed by considering the hardware limitations, while an adaptive mixed-precision quantization module is developed to evaluate the quantization sensitivity by using the Hessian matrix and Pareto frontier techniques. Integer linear programming is used to fine-tune the quantization across different layers. Then the low-cost proxy neural architecture search module efficiently explores the ideal quantization hyperparameters. Experiments on the ImageNet demonstrate that the proposed LCPAQ achieves comparable or superior quantization accuracy to existing mixed-precision models. Notably, LCPAQ achieves 1/200 of the search time compared with existing methods, which provides a shortcut in practical quantization use for resource-limited devices.