自适应调整梯度采样数，实现INT8芯片上高效零阶训练
QScheduler: Adaptive Gradient Sampling for Zeroth-Order On-Device Training on INT8 NPUs
- 根据训练进度动态调节梯度采样次数，减少人工调参
- 在STM32N6芯片上首次实现INT8量化端侧训练，性能媲美固定采样
- 适用于资源受限设备的模型微调，尤其适合嵌入式AI场景
零阶优化通过仅使用前向传播估算梯度，使配备NPU的微控制器实现端侧学习，无需反向传播且降低内存占用。梯度采样次数q对训练效果至关重要：采样不足导致梯度噪声大、过早收敛；采样过多则增加计算开销。然而，寻找最优q通常需代价高昂的超参数搜索。本文提出QScheduler，一种基于训练进度自适应调整q的算法，并首次在STM32N6的Neural-ART NPU上实现INT8量化端侧训练。在EuroSAT和STL-10数据集上的实验表明，QScheduler在ResNet18与MobileNetV2上表现与经过良好调优的固定q配置相当，且无需预先进行q的超参数优化。
原文摘要 · Abstract (English)
Zeroth-Order (ZO) optimization enables On-Device Learning (ODL) on NPU-equipped microcontrollers by estimating gradients through forward passes alone, bypassing the need for backpropagation primitives and reducing memory requirements. The number of gradient samples q critically affects training: insufficient samples produce noisy gradients that plateau early, while excessive samples consume more computational resources. However, finding an optimal q typically requires costly hyperparameter searches. This work introduces QScheduler, an adaptive algorithm that adjusts q based on training progress, and provides the first proof-of-concept of INT8 quantized on-device training on the STM32N6's Neural-ART NPU. Experiments on EuroSAT and STL-10 show that QScheduler matches well-tuned fixed-q configurations for both ResNet18 and MobileNetV2, without requiring prior q hyperparameter optimization.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。