On PAC Learning Using Winnow, Perceptron, and a Perceptron-like Algorithm