arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 1904.01705cs.ETcs.NE
收藏

提高混合信号神经网络的抗噪能力

Improving Noise Tolerance of Mixed-Signal Neural Networks

Michael Klachko, Mohammad Reza Mahmoodi, Dmitri B. Strukov

摘要:

用于深度学习的混合信号硬件加速器的能效比数字硬件加速器高出几个数量级。在超低功耗状态下,模拟计算固有的有限信号精度成为一个挑战。我们进行了一个案例研究的6层卷积神经网络上运行的混合信号加速器,并评估其对硬件特定噪声的敏感性。我们应用各种方法来提高网络的噪声鲁棒性,并展示了一种有效的方法来优化有用的信号范围,通过自适应信号限幅。由此产生的模型足够强大,仅用1.4 mW的功率预算就可以在CIFAR-10数据集上实现80.2%的分类准确率,而6 mW的预算使我们能够实现87.1%的准确率,这在软件基线的1%以内。相比之下,相同模型的未优化版本在1.4 mW和6 mW时分别仅达到67.7%和78.6%的准确度。

英文摘要:

Mixed-signal hardware accelerators for deep learning achieve orders of magnitude better power efficiency than their digital counterparts. In the ultra-low power consumption regime, limited signal precision inherent to analog computation becomes a challenge. We perform a case study of a 6-layer convolutional neural network running on a mixed-signal accelerator and evaluate its sensitivity to hardware specific noise. We apply various methods to improve noise robustness of the network and demonstrate an effective way to optimize useful signal ranges through adaptive signal clipping. The resulting model is robust enough to achieve 80.2% classification accuracy on CIFAR-10 dataset with just 1.4 mW power budget, while 6 mW budget allows us to achieve 87.1% accuracy, which is within 1% of the software baseline. For comparison, the unoptimized version of the same model achieves only 67.7% accuracy at 1.4 mW and 78.6% at 6 mW.

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。

↑