Application Model. Figure 5.1: Adaptive dynamic data driven framework for uncertain spatial task as- signment
Application Model. We follow the application model presented in Section 2.1.3. Namely, the target model of the attack is a binary classification model that predicts whether an input sample belongs to the benign or malware class. Recall that to construct such a classifier for malware detection, executables are represented as binary feature vectors. for this purpose, datasets often provide a comprehensive set of extracted features from real-world executables. With these datasets and features 1 . . . M, we can construct a vector X for each input sample such that X ∈ {0, 1}M . Xi = 1 indicates the presence of feature i and Xi = 0 indicates its absence. We use the feature vectors and associated class labels to construct binary classification models for malware detection, similar to the one shown in figure 2.1.
