SDS 551: Deep Reinforcement Learning — with Wah Loon Keng

RL Strategies

The evolution of reinforcement learning (RL) strategies is marked by significant milestones, such as the development of AlphaGo Zero and AlphaZero. Wah Loon Keng highlights how AlphaGo Zero learned without human data, a leap from its predecessor, AlphaGo, which relied on human gameplay. AlphaZero further advanced this by mastering multiple games like Go, chess, and shogi without human input 1.

AlphaZero is no human knowledge, but it masters go chess and another game called shogi.

--- Wah Loon Keng

The discussion also touches on the stagnation in new RL algorithms, with the last surge of innovations occurring in 2017. However, Keng notes that DeepMind's work on open-ended play and emergent behavior offers promising directions for overcoming current limitations 2.

   

Implications

The long-term implications of reinforcement learning advances are profound, impacting both technology and society. Jon Krohn explores how deep RL is currently applied in industries and speculates on its future potential. These advancements could revolutionize various sectors, enhancing automation and decision-making processes 3.

Very cool to hear about how deep reinforcement learning is being used in industry today.

--- Jon Krohn

As RL continues to evolve, it promises exciting developments in AI, potentially transforming how we interact with technology and solve complex problems 4.