Norm-Preservation: Why Residual Networks Can Become Extremely Deep?
Sign Up to like & getrecommendations! Published in 2021 at "IEEE Transactions on Pattern Analysis and Machine Intelligence"
DOI: 10.1109/tpami.2020.2990339
Abstract: Augmenting neural networks with skip connections, as introduced in the so-called ResNet architecture, surprised the community by enabling the training of networks of more than 1,000 layers with significant performance gains. This paper deciphers ResNet… read more here.
Keywords: preservation; norm preservation; residual networks; networks become ... See more keywords