Abstract: In multi-instance learning, the training set comprises labeled bags that are composed of unlabeled instances, and the task is topredict the labels of unseen bags. This paper studies multi-instancelearning from the view of supervised learning. First, by analyzing somerepresentative learning algorithms, this paper shows that multi-instancelearners can be derived from supervised learners by shifting theirfocuses from the discrimination on the instances to the discriminationon the bags. Second, considering that ensemble learning paradigms caneffectively enhance supervised learners, this paper proposes to buildmulti-instance ensembles to solve multi-instance problems. Experimentson a real-world benchmark test show that ensemble learning paradigms cansignificantly enhance multi-instance learners.