Deep residual networks have emerged as a family of extremely deep
architectures showing compelling accuracy and nice convergence behaviors. In
this paper, we analyze the propagation formulations behind the residual
building blocks, which suggest that the forward and backward signals can be
directly propagated from one block to any other block, when using identity
mappings as the skip connections and after-addition activation. A series of
ablation experiments support the importance of these identity mappings. This
motivates us to propose a new residual unit, which makes training easier and
improves generalization. We report improved results using a 1001-layer ResNet
on CIFAR-10 (4.62% error) and CIFAR-100, and a 200-layer ResNet on ImageNet.
Code is available at: https://github.com/KaimingHe/resnet-1k-layers
Description
[1603.05027] Identity Mappings in Deep Residual Networks
%0 Generic
%1 he2016identity
%A He, Kaiming
%A Zhang, Xiangyu
%A Ren, Shaoqing
%A Sun, Jian
%D 2016
%K 2016 computer-vision deep-learning microsoft
%T Identity Mappings in Deep Residual Networks
%U http://arxiv.org/abs/1603.05027
%X Deep residual networks have emerged as a family of extremely deep
architectures showing compelling accuracy and nice convergence behaviors. In
this paper, we analyze the propagation formulations behind the residual
building blocks, which suggest that the forward and backward signals can be
directly propagated from one block to any other block, when using identity
mappings as the skip connections and after-addition activation. A series of
ablation experiments support the importance of these identity mappings. This
motivates us to propose a new residual unit, which makes training easier and
improves generalization. We report improved results using a 1001-layer ResNet
on CIFAR-10 (4.62% error) and CIFAR-100, and a 200-layer ResNet on ImageNet.
Code is available at: https://github.com/KaimingHe/resnet-1k-layers
@misc{he2016identity,
abstract = {Deep residual networks have emerged as a family of extremely deep
architectures showing compelling accuracy and nice convergence behaviors. In
this paper, we analyze the propagation formulations behind the residual
building blocks, which suggest that the forward and backward signals can be
directly propagated from one block to any other block, when using identity
mappings as the skip connections and after-addition activation. A series of
ablation experiments support the importance of these identity mappings. This
motivates us to propose a new residual unit, which makes training easier and
improves generalization. We report improved results using a 1001-layer ResNet
on CIFAR-10 (4.62% error) and CIFAR-100, and a 200-layer ResNet on ImageNet.
Code is available at: https://github.com/KaimingHe/resnet-1k-layers},
added-at = {2018-05-06T13:10:25.000+0200},
author = {He, Kaiming and Zhang, Xiangyu and Ren, Shaoqing and Sun, Jian},
biburl = {https://www.bibsonomy.org/bibtex/2f084b87b38f85def5512128fb304dba4/achakraborty},
description = {[1603.05027] Identity Mappings in Deep Residual Networks},
interhash = {bcf33120cfac129f48c2c50673214869},
intrahash = {f084b87b38f85def5512128fb304dba4},
keywords = {2016 computer-vision deep-learning microsoft},
note = {cite arxiv:1603.05027Comment: ECCV 2016 camera-ready},
timestamp = {2018-05-06T13:10:25.000+0200},
title = {Identity Mappings in Deep Residual Networks},
url = {http://arxiv.org/abs/1603.05027},
year = 2016
}