Mileage: Multiple instance learning with global embedding

Multiple Instance Learning (MIL) generally rep-resents each example as a collection of instances such that the features for local objects can be bet-ter captured, whereas traditional learning meth-ods typically extract a global feature vector for each example as an integral part. However, there is limited research work on investigating which of the two learning scenarios performs better. This paper proposes a novel framework – Mul-tiple Instance LEArning with Global Embedding (MILEAGE), in which the global feature vectors for traditional learning methods are integrated into the MIL setting. MILEAGE can leverage the benefits derived from both learning settings. Within the proposed framework, a large mar-gin method is formulated to adaptively tunes the weights on the two different kinds of feature rep-resentations (i.e., global and multiple instance) for each example and trains the classifier simul-taneously. Some important properties of the pro-posed method are analyzed. Experiments on im-age, text and a novel application – Insider Threat Detection are conducted to demonstrate the ad-vantages of the proposed method. 1