Multiple instance real boosting with aggregation functions

We introduce a boosting framework for multiple instance learning (MIL) with varied aggregation of instances. In this framework, a diverse set of aggregation functions can be used to refine the notion of a positive bag for multiple instance learning. We investigate the effect of a wide range of orness in aggregation, using ordered weighted averaging. Thus, we obtain a new notion of a positive bag, which can represent different levels of ambiguity. We evaluate the performance of the proposed algorithm on popular MIL datasets. The experimental results show that this algorithm outperforms the standard MILBoost algorithm.