High-precision linearized interpretation for fully connected neural network

Open Webpage

Xia Lei, Yongkai Fan, Kuan-Ching Li, Arcangelo Castiglione, Qian Hu

Published: 01 Jan 2021, Last Modified: 10 Apr 2025Appl. Soft Comput. 2021EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Highlights•Converting deep neural networks into a set of mathematically convergent piecewise linear neural networks.•Equivalent linear classifiers for piecewise linear neural networks.•Convergence and consistency of CIDNN’s interpretation model.•Semantical meaning of CIDNN in the Fashion-MNIST dataset.