In this section we present the results of a controlled experiment designed to estimate the performance of the proposed method when deployed in the real world. We envision two applications of the method: 1) malware detection, i.e., identify whether a new, unknown app is malware or benign, and 2) familial classification,