THE ALGORITHM: THE PERCEPTRON UPDATE RULE (This rule and proof adapted from Weinberger’s lecture.) Step 1. Initialize the weight vector to zero: set w = 0. Step 2. For each data point x in the training dataset, do the following: Step 2a if ywTx ≤ 0: the weight vector is wrong, so update it: wnew= wold + yx Step 3. If there were no updates to the weight vector in step 2, terminate, otherwise go to step 2 and iterate over all the data points once again.
