On an Interpretation of ResNets via Solution Constructions
Machine Learning
2022-12-27 v2 Artificial Intelligence
Machine Learning
Abstract
This paper first constructs a typical solution of ResNets for multi-category classifications by the principle of gate-network controls and deep-layer classifications, from which a general interpretation of the ResNet architecture is given and the performance mechanism is explained. We then use more solutions to further demonstrate the generality of that interpretation. The universal-approximation capability of ResNets is proved.
Cite
@article{arxiv.2212.05663,
title = {On an Interpretation of ResNets via Solution Constructions},
author = {Changcun Huang},
journal= {arXiv preprint arXiv:2212.05663},
year = {2022}
}
Comments
v2:writing improved