Consider designing a linear classifier
on a dataset , , . Recall that the sign function outputs +1 if the argument is positive, and −1 if the argument is non-positive. The parameters and are updated as per the following training algorithm:
whenever . In other words, whenever the classifier wrongly predicts a sample from the dataset, gets updated to , and likewise gets updated to . Consider the case , . Then
- A
- B
- C
- D

