Anti-Martingale Proximal Policy Optimization