learning from label proportions (LLP) is a promising weakly supervised
learning problem. In LLP, a set of instances (bag) has label proportions, but
no instance-level labels are given. LLP aims to train an instance-level
classifier by using the label proportions of the bag. In this pap