Hard Negative Mining

The hard negative examples are easily misclassified. We can explicitly find those false positive samples during the training loops and include them in the training data so as to improve the classifier.