Let T denote the size of the training set, the binary feature vector of the ith pair in the training set be denoted by X i, y i = 1 if the pair is annotated as positive and y i = −1 otherwise, w denote a vector of feature weights, of the same length as X i, θ denote a threshold parameter and λ denote a regularization parameter.
In the first approach we added to the binary feature vector coding the given sequence for the SVM the description of its location in sequence space.
As its name implies, this conditional probability function is a maximum-entropy distribution, constrained to conform with the binary feature frequencies observed in the training data.
However, the forms of such binary feature descriptors are predefined in the hand-crafted way, which requires strong domain knowledge to design them.
Dictionary features included binary features that indicate the presence of the word in our dictionary and the position of the word within any dictionary entries.
Here (x_{ij}in {0,1}) is the binary feature of (G_j) corresponding to the subgraph pattern (g_i), and (x_{ij}=1) iff (g_i) is a subgraph of (G_j) ((g_isubseteq G_j)), otherwise (x_{ij}=0).
Section 'Concave-convex local binary features' discusses the limitation of traditional LBP features and presents the new CCP scheme in detail.
Finally, in this work, we are only concerned with patterns represented by binary feature vectors x residing in binary feature space: x (in mathcal{{X}} ={0,1}^{L}), i.e., (x_{l}in ~{0,1}).
In this respect, CAC outperforms SVM with the best kernel for binary features, and provides the promising results equivalent to SVM on average for non-binary features.
