On an Interpretation of ResNets via Solution Constructions
This provides a theoretical foundation for understanding ResNets, which is incremental as it builds on existing architectures.
The paper tackles the problem of interpreting ResNet architectures by constructing solutions for multi-category classification, explaining their performance mechanism, and proving their universal-approximation capability.
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.