Maximize
set of misclassified samples under w only update until convergence
Weight Updates
Start with weight 0
For each training instance
- If correct (i.e., y=y*), no change
- If wrong: adjust the weight vector - adding or subtracting the feature vector (f / y* exactly
x0 bias ican be tan add to f
Error-Driven Linear Classification
Binary case - compare features to weight vector Learning - figure out weight vector from examples