正线性函数在深度神经网络中的研究

摘要

Aiming at the problems of non‐linearity and high computational complexity of the common activation functions in the deep neural networks ,the method using the positive linear function to replace the common non‐linear activation function in deep neural networks was presented .Positive linear function not only has low computational complexity ,but also implements the sparse representation of information like human brain .The image classification task was used to verify the effectiveness of the positive linear function in deep neural networks .%针对深度神经网络中常用的激活函数具有非线性和计算复杂度高的特点，提出使用正线性函数代替常用非线性函数作为深度神经网络的激活函数。基于正线性激活函数建立的深度神经网络模型计算复杂度低，能得到稀疏表示，与人类大脑信息感知具有一致性。通过图像分类任务验证了正线性激活函数在深度神经网络中应用的有效性。