Active Learning with Partial Feedback

It is expensive to obtain full information from labeler: e.g. picking fine-grained category. Can the algorithm actively decide what to ask labeler? Can we move on to a different example in a strategic way?

For more details, visit the source.