Algorithm 1 Learning Rate Update Algorithm

1. Input: Learning rate lr; minimum learning rate min_lr; initial learning rate init_lr; maximum number of iterations τ; warm-up iterations τ * 2. β = 1 3. while Termination conditions are not met, do 4. if β < τ * 5.  One-dimensional linear interpolation g = [0,τ *] 6.  Update lr = g 7.  β = τ * 8.  else 9.  Update lr = min_lr + (init_lr-min_lr)*((1 + cos($\pi \frac{\beta }{\tau }$))/2) 10.  end if 11.  β = β + 1 12.  end while 13.  Output: Update learning rate lr *