Learning from Loss Landscape: Generalizable Mixed-Precision Quantization via Adaptive Sharpness-Aware Gradient Aligning

TL;DR

This paper introduces a novel mixed-precision quantization method that generalizes policies from small to large datasets using sharpness-aware optimization, reducing computational costs and maintaining accuracy.

Contribution

The paper proposes a new approach combining sharpness-aware minimization and adaptive gradient alignment to efficiently transfer quantization policies from small to large datasets.

Findings

Achieved ImageNet accuracy with 0.5% of training data

Reduced computational cost by up to 150% compared to baselines

Validated effectiveness through theoretical analysis and experiments

Abstract

Mixed Precision Quantization (MPQ) has become an essential technique for optimizing neural network by determining the optimal bitwidth per layer. Existing MPQ methods, however, face a major hurdle: they require a computationally expensive search for quantization policies on large-scale datasets. To resolve this issue, we introduce a novel approach that first searches for quantization policies on small datasets and then generalizes them to large-scale datasets. This approach simplifies the process, eliminating the need for large-scale quantization fine-tuning and only necessitating model weight adjustment. Our method is characterized by three key techniques: sharpness-aware minimization for enhanced quantization generalization, implicit gradient direction alignment to handle gradient conflicts among different optimization objectives, and an adaptive perturbation radius to accelerate optimization. Both theoretical analysis and experimental results validate our approach. Using the CIFAR10 dataset (just 0.5\% the size of ImageNet training data) for MPQ policy search, we achieved equivalent accuracy on ImageNet with a significantly lower computational cost, while improving efficiency by up to 150% over the baselines.