Learning operates on feature selection space, which is significantly smaller than the traditional architectural-level configuration space. The features in FUSION encode the engineer’s domain knowledge of the practical variation points in a given application. For instance, the engineer may only consider a small reasonable subset of MNP authentication driven architectural choices (recall Section 2). Figure 1b shows two authentication strategies modeled as features in TRS: F3 and F4. These two features represent what the TRS security engineer envisioned to be the reasonable applications of authentication in the system.