Active Learning with Deep Learning

I'm amazed by how little this topic is dicussed in the industry. Roughly speaking:

You have a stack of unlabeled images and a fixed labeling budget. Which images should You send for labeling?

This is at least as important as the type of network you use. In fact the only thing that's more important when doing practical projects is the problem formulation itself.

Anyway, on to the links:

http://jacobgil.github.io/deeplearning/activelearning

https://towardsdatascience.com/review-suggestive-annotation-deep-active-learning-framework-biomedical-image-segmentation-e08e4b931ea6