Research Article

[Retracted] Research on Network Layer Recursive Reduction Model Compression for Image Recognition

Algorithm 1

Our method.
Initialization: learning rate , number of residual units , number of residual units removed at one time is , number of retraining iterations is .
(l)ResNet’s parameters and weights are initialized.
(2)Set the initial learning rate to train ResNet.
(3)
(4)while do
(5)
(6)Select in ascending order of the absolute value of and add that index to
(7)Delete of from equation (2)
(8)
(9)Retraining in SGD with initial learning rate for iterations
(10)End