[Retracted] Gradient Descent Optimization in Deep Learning Model Training Based on Multistage and Method Combination Strategy

<div>Performance of decay and combination of cosine decay and decay during 100 epochs of training on MNIST. (a, b) The accuracy and loss of the result (init-lr = 0.01).</div>

Security and Communication Networks

fig6

Figure 6

Figure 6: [Retracted] Gradient Descent Optimization in Deep Learning Model Training Based on Multistage and Method Combination Strategy 