VOVU: A Method for Predicting Generalization in Deep Neural Networks

Figure 4

The dynamic trend of the test loss and VOVU trained on CIFAR-10 in convolutional neural networks. The first part of the VOVU curve does not decrease much, but the point when test loss starts to rise is the same point VOVU starts to rise (the learning rate is set to 0.0008, 0.001, and 0.0012 from left to right, respectively). (a) Learning rate = 0.0008. (b) Learning rate = 0.001. (c) Learning rate = 0.0012.

(a)

(b)

(c)