Negative Update Intervals in Deep Multi-Agent Reinforcement Learning