浅析ResNet50

Brief Analysis of ResNet50

摘要: 针对深层次的神经网络训练问题，提出了简化对更深网络的训练，提升速度和精度的残差网络(ResNet)，ResNet作为卷积神经网络的一个里程碑式的模型。ResNet50为ResNet的一个代表，其残余网络为了解决梯度消失，深层网络训练的问题，使用了残差连接提起残差特征，作为该层输入中学习特征的减法，使得其在各个领域的应用更为广泛。

Abstract: Aiming at the problem of deep neural network training, this paper proposes a residual network (ResNet), which simplifies the training of deeper networks and improves the speed and accuracy. ResNet is a landmark model of convolutional neural networks. ResNet50 is a representative of ResNet. In order to solve the problems of gradient disappearance and deep network training, its residual network uses residual connection to lift residual features as the subtraction method of learning features in the input of this layer, which makes it more widely used in various fields.