Harri Valpola: System 2 AI and Planning in Model-Based Reinforcement Learning
Published:May 25, 2020 11:00
•1 min read
•ML Street Talk Pod
Analysis
This article summarizes a podcast episode discussing System 1/2 thinking in AI, model-based reinforcement learning (RL), and related research. It highlights the challenges of applying model-based RL to industrial control processes and introduces a recent paper by Curious AI on regularizing trajectory optimization. The episode covers various aspects of the topic, including the source of simulators, evolutionary priors, consciousness, company building, and specific techniques like Deep Q Networks and denoising autoencoders. The focus is on the practical application and research advancements in model-based RL.
Key Takeaways
- •The podcast episode discusses System 1/2 thinking in AI and its application to model-based reinforcement learning.
- •Challenges of applying model-based RL to industrial control processes are highlighted.
- •A recent paper by Curious AI on regularizing trajectory optimization with denoising autoencoders is introduced.
- •The episode covers various aspects of model-based RL, including simulators, evolutionary priors, and specific techniques like Deep Q Networks.
Reference
“Dr. Valpola and his collaborators recently published “Regularizing Trajectory Optimization with Denoising Autoencoders” that addresses some of the concerns of planning algorithms that exploit inaccuracies in their world models!”