multiple-instance learning (mil)

A form of weakly supervised learning where training instances are arranged in sets, or bags, with labels assigned to bags instead of individual instances.