[Lesson 7: Resnet-ish CNN ] Why do the loss graphs look like this?

Hello everybody!

After training the last, Resnet-ish CNN I plotted the losses and metrics with:

learn.recorder.plot_losses()
learn.recorder.plot_metrics()

And got the following graphs:

Why did the validation loss spike around epoch number 2000?