通过改进Inception模块，分解卷积，提出了Inception V3
Residual Networks Behave Like Ensembles of Relatively Shallow Networks
Residual connections使得网络变得更深，但作者从Ensembles角度来解释Residual Networks性能提升原因
Batch Normalization--Accelerating Deep Network Training by Reducing Internal Covariate Shift
减少层的“covariate shift”，提出Batch Normalization算法，大大改善了网络训练过程
Understanding the difficulty of training deep feedforward neural networks
讲解神经网络训练难得原因，提出Xavier权重初始化方法
ILSVRC历届冠军论文笔记
ImageNet Large Scale Visual Recognition Challenge是图像分类领域的比赛，记录一下历届冠军/经典论文的笔记。