Paper ID: 2212.05663

On an Interpretation of ResNets via Solution Constructions

Changcun Huang

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.

Submitted: Dec 12, 2022