Each such iteration is referred to as a boosting iteration.
In the TrAdaBoost algorithm, during every boosting iteration, a so-called weak classifier is built using weighted instance data from the previous iteration.
Assuming there are a total of N boosting iteration rounds, in the round n, we will build K weak classifiers ([h_{i}^{(n)}(mathbb {F}_{k})]_{k=1}^{K}) for each local patch in parallel from the boosting sample subset.
Specifically, at each boosting iteration, a regularized Operator-valued Kernel-based Vector AutoRegressive model (OKVAR) is trained on a random subnetwork.
The final prediction is in the form of a linear combination of stump classifier predictions (one per boosting iteration, where the weight of each stump classifier is determined by its error rate).
Since boosting iteration adds three nodes (one splitter node and two prediction nodes) to the tree, more boosting iterations will result in larger and potentially more accurate trees.
Therefore, for each boosting iteration,, iteration's specific weak classifier,, with the minimum error,, is selected.
In this case each bagging or boosting iteration contributes independently to the SOHN.
At each boosting iteration, several weak classifiers are learned (one for each subwindow), and the best one is kept.
At each boosting iteration the weights are adjusted through the use of (17), resulting in the following recursive relation: (18).
Assuming, in each boosting iteration stage, that there are K possible local patches, which are represented by SURF feature (mathbb {F}), each stage is a boosting training procedure with logistic regression as weak classifiers.
