基于双阈值的ANN-SNN转换方法优化  

Optimization of ANN-SNN Conversion Method with Double Threshold

作　　者：何赟泽 张天安 邓堡元 王洪金 王耀南[1,2] HE Yunze;ZHANG Tianan;DENG Baoyuan;WANG Hongjin;WANG Yaonan(School of Electrical and Information Engineering,Hunan University,Changsha 410082,China;Robot Visual Perception and Control Technology of National Engineering Reasearch Center,Changsha 410082,China)

机构地区：[1]湖南大学电气与信息工程学院,长沙410082 [2]机器人视觉感知与控制技术国家工程研究中心,长沙410082

出　　处：《计算机测量与控制》2024年第11期271-277,共7页Computer Measurement &Control

基　　金：国家自然科学基金(62101184);湖南省科技创新领军人才(2023RC1039);湖南省自然科学基金重大项目(2021JC0004)。

摘　　要：脉冲神经网络作为第三代神经网络,能够克服许多人工神经网络中所存在的问题,如高功耗、鲁棒性较差等;通过对预训练好的人工神经网络模型进行转换是获取深度脉冲神经网络模型的一种主要方法,然而通过这种方法获取的脉冲神经网络的延迟较高,无法满足实时性要求;文章在双阈值转换方法的基础上,采用阈值平衡技术对转换过程进行优化,通过理论推导,提出了一种对称阈值LeakyReLU激活函数,并对人工神经网络到脉冲神经网络的转换流程进行了梳理;此外,采用了泄漏机制对转换后的脉冲神经网络模型结构进行了优化,并通过脉冲时序依赖可塑性学习规则对该结构进行训练;最终,在MNIST数据集与CIFAR-10数据集上进行了实验,结果表明,优化后脉冲神经网络的收敛速度与鲁棒性得到了大幅提升。As a third-generation neural network,spiking neural network(SNN)can overcome many problems in artificial neural networks,such as high power consumption and poor robustness.Transforming the pre-trained artificial neural network(ANN)model is a primary method for obtaining deep spiking neural network model,but this spiking neural network has a high latency and cannot meet real-time requirements.On the basis of the double threshold conversion method,the threshold balance technology is used to optimize the transition process.Through theoretical derivation,a symmetric threshold LeakyReLU activation function is proposed,and the transition process from artificial neural network to spiking neural network is sorted out.In addition,the leakage mechanism is used to optimize the structure of the transformed spiking neural network model,and the structure is trained by the spiking-timing-dependent plasticity learning rule.Finally,the experiments are carried out on the MNIST dataset and the CIFAR-10 dataset,and the results show that the optimized spiking neural network greatly improves the convergence speed and robustness.

关 键 词：ANN-SNN转换 双阈值 阈值平衡 脉冲时序依赖可塑性 泄漏机制 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]