TCAV for multiple instance learning

[qinghezeng]

Hi, many thanks for this great work!

I wonder does it make sense to use TCAV to test an attention based multiple instance network? If yes, the examples of a concept should be bags or instances?

Thank you very much!