Discovering Diverse Solutions in Deep Reinforcement Learning by Maximizing State-Action-Based Mutual Information | Read Paper on Bytez