5. Recall that the conversion from an online algorithm with mistake bound m to a PAC algorithm given in class works as follows: "Run A on a sequence of examples each drawn independently from D : If hypothesis h ever survives (1/)log(m+1) consecutive examples without making a mistake, stop and output h." Now suppose that you have an online algorithm A with some finite mistake bound m, but you don't know what the value of m is. Explain how you can obtain a PAC algorithm from A. What is the best sample complexity (in terms of m,, and ) that you can achieve for your PAC algorithm