Paper proposes a new model using consistency regularization for learning from label proportions.
problem Learning from label proportions with weak labels on bags of instances.
method Consistency regularization applied to semi-supervised learning.
result LLP with consistency regularization achieves superior performance.