Quasar Model-Free Q-Learning Reduces Memory Footprint for MDPs

Quasar is a model-free RL algorithm providing asymptotic guarantees for reachability in MDPs without non-terminal maximal end components. It utilizes temporal-difference updates to reduce memory requirements from O(|S|^2|A|) to O(|S||A|) compared to model-based methods.