I would like to ask about the state of the art of learning from a bag of images labeled with one label. What are the architectures of CNN that handle several images having only one label to do image classification? Please share with me your thoughts.