深度单峰梯形神经网络  

Deep single-peaked trapezoid neural networks

作　　者：单传辉 SHAN Chuanhui(College of Computer Science,Faculty of Information Technology,Beijing University of Technology,Beijing 100124,China)

机构地区：[1]北京工业大学信息学部计算机学院,北京100124

出　　处：《计算机工程与应用》2018年第23期7-13,共7页Computer Engineering and Applications

基　　金：国家自然科学基金(No.61175004)

摘　　要：近年来,深度学习在许多领域取得了巨大的成功,其中,激活函数是深度学习取得巨大成功的关键因素之一。根据生物神经元特性,针对ReLU右侧响应无界问题,提出了单峰梯形线性单元(Single-Peaked Trapezoid Linear Unit,SPTLU)。SPTLU更加符合生物神经元特性,且取得了等同和超越ReLU的优异性能,实验表明在不同数据集上都取得了很好的效果,例如,数据集MNIST,Fashion-MNIST,SVHN,CALTECH101和CIFAR10。Recently,deep learning has obtained a great success in many research areas.The activation function is one of the key factors in deep learning.In this paper,according to characteristics of biological neurons,an improved Single-Peaked Trapezoid Linear Unit(SPTLU)activation function is presented for the right-hand response unbounded of ReLU.SPTLU is in line with the biological neuron biological essence and achieves the excellent performance of equivalent or beyond ReLU.The experiments show the proposed activation function achieves good effectiveness on different datsets,e.g.,MNIST,Fashion-MNIST,SVHN,CALTECH101 and CIFAR10 datasets.

关 键 词：单峰梯形线性单元(SPTLU) SPTLU神经元 SPTLU网络 对比实验 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]