Figure 4
From: Enhancing deep neural network training efficiency and performance through linear prediction
Loss and comparison on validation set of proposed PLP method, DEMON and SGD. The legend “XXX-XXX” in “Loss Diff” refers to the curve shown in figure is the value of “loss of XXX method subtract XXX’s loss”. The black horizontal solid line is the line of zero, and the red horizontal dashed line represents the average value of the loss difference.
