Overfitting yet Validation loss decreasing and Accuracy increasing - what does it mean?

Deep Learning
1

Hi students. I understand that Training loss decreasing relative to Validation indicates overfitting. Which I interpret intuitively as the model memorizing training examples and their classes.

But what if Validation loss and Accuracy continue to improve at the same time?

What I’d like to know:

  • Intuitively, in this situation, what is going on for the model and training?

  • How to respond to it in practice?

Thanks for your insights.

1 Like