Optimal Weak to Strong Learning