Planning to Be Surprised: Optimal Bayesian Exploration in Dynamic Environments

Abstract

To maximize its success, an AGI typically needs to explore its initially unknown world. Is there an optimal way of doing so? Here we derive an affirmative answer for a broad class of environments.

Cited by in corpus (8)

• Graph networks as learnable physics engines for inference and control
• Surprise-Based Intrinsic Motivation for Deep Reinforcement Learning
• Reinforcement Learning through Active Inference
• Is Pessimism Provably Efficient for Offline RL?
• POWERPLAY: Training an Increasingly General Problem Solver by Continually Searching for the Simplest Still Unsolvable Problem
• Learning-Driven Exploration for Reinforcement Learning
• Tutorial and Survey on Probabilistic Graphical Model and Variational Inference in Deep Reinforcement Learning
• A Bandit Framework for Optimal Selection of Reinforcement Learning Agents