Overview
Save Big on Coursera Plus. 7,000+ courses at $160 off. Limited Time Only!
Explore a groundbreaking reinforcement learning algorithm called Go-Explore that tackles hard-exploration problems in Atari games. Learn how this modern adaptation of Dijkstra's shortest path algorithm outperforms existing methods by orders of magnitude through random exploration. Discover the key principles behind Go-Explore, including state memory, targeted exploration, and robustification through imitation learning. Examine its impressive performance on notoriously difficult games like Montezuma's Revenge and Pitfall, where it achieves superhuman scores and surpasses previous state-of-the-art results. Gain insights into the algorithm's potential applications in various domains, including robotics, and its implications for future research in reinforcement learning and intelligent exploration.
Syllabus
Go-Explore: a New Approach for Hard-Exploration Problems
Taught by
Yannic Kilcher