Fast Two-Time-Scale Stochastic Gradient Method with Applications in Reinforcement Learning