arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 1502.02551cs.LGcs.NEstat.ML
收藏

数值精度有限的深度学习

Deep Learning with Limited Numerical Precision

IBM T. J. Watson Research Center(IBM T·J·沃森研究中心);IBM Almaden Research Center(IBM阿尔马登研究中心)

Suyog Gupta, Ankur Agrawal, Kailash Gopalakrishnan, Pritish Narayanan

摘要:

大规模深度神经网络的训练通常受到可用计算资源的约束。研究了有限精度的数据表示和计算对神经网络训练的影响。在低精度定点计算的背景下,我们观察到舍入方案在确定网络在训练过程中的行为方面起着至关重要的作用。我们的研究结果表明,当使用随机舍入时,深度网络可以仅使用16位宽的固定点数表示进行训练,并且分类精度几乎没有下降。我们还展示了一个节能的硬件加速器,实现低精度定点运算随机舍入。

英文摘要:

Training of large-scale deep neural networks is often constrained by the available computational resources. We study the effect of limited precision data representation and computation on neural network training. Within the context of low-precision fixed-point computations, we observe the rounding scheme to play a crucial role in determining the network's behavior during training. Our results show that deep networks can be trained using only 16-bit wide fixed-point number representation when using stochastic rounding, and incur little to no degradation in the classification accuracy. We also demonstrate an energy-efficient hardware accelerator that implements low-precision fixed-point arithmetic with stochastic rounding.

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。

↑