arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2610.17099cs.ARcs.LG
收藏

LocalTrain:面向边缘FPGA的节能CNN训练的纯整数本地学习

LocalTrain: Integer-Only Local Learning for Energy-Efficient CNN Training on Edge FPGAs

University of Cambridge(剑桥大学);University College London(伦敦大学学院);University of Bristol(布里斯托大学)

Shiqi Ding, Xuanyu Liang, Xinyang Liu, Boyang Li, Zhiqiang Que, Guosheng Hu, Jiayang Li

AI总结:

提出面向边缘FPGA的LocalTrain纯整数本地学习系统,用共享整数矩阵引擎训练CNN,在KV260上实现高准确率,能耗较Jetson Orin Nano同类方法显著降低。

摘要:

本地学习可减少跨层训练依赖,但高效加速器仍需保留操作数、累积梯度并选择量化尺度。本文提出一种FPGA系统,使用共享整数矩阵引擎训练CNN及其区域辅助分类器;片上扩展紧凑误差,逐块折叠梯度,历史尺度支持直接输出转换。在200 MHz下,KV260原型在CIFAR-10上准确率达76.57%,在SVHN上达91.85%,平均模块功耗约4.20 W,有效矩阵吞吐量分别为31.22 GMAC/s和31.57 GMAC/s。与Jetson Orin Nano上的同类整数方法相比,训练能耗分别降低40.4%和42.1%,运行时间分别延长1.86倍和1.82倍;测量采用各平台电源域,所选状态检查及完整运行预测计数与软件参考功能一致。

英文摘要:

Local learning reduces cross-layer training dependencies, but an efficient accelerator must still retain operands, accumulate gradients, and select quantization scales. We present an FPGA system that trains a CNN and its regional auxiliary classifiers using a shared integer matrix engine. Compact errors are expanded on chip, gradients are folded tile by tile, and historical scales permit direct output conversion. At 200 MHz, the KV260 prototype achieves 76.57% accuracy on CIFAR-10 and 91.85% on SVHN at approximately 4.20 W mean module power. Effective matrix throughput is 31.22 and 31.57 GMAC/s, respectively. Against the same integer method on Jetson Orin Nano, measured training energy is 40.4% and 42.1% lower, with $1.86\times$ and $1.82\times$ longer runtime; measurements use the respective platform power domains. Selected state checks and matching complete-run prediction counts support functional agreement with the software reference.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑