Intuition for Perceptron Training Rule

• If the training example is correctly classified , making , so no weights are updated
• If the perceptron outputs -1 when the target output is +1 and assuming and , then
• if the perceptron outputs +1 when the target output is -1, then the weight would be decreased
• this learning procedure will converge within finite number of applications of the perceptron training rule to a weight vector that correctly classifies all training examples, provided the training examples are linearly separableБе and a sufficiently small is used.