Reproduce “Neuro-Symbolic World Models for Adapting to Open World Novelty” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jonathan Balloch, Zhiyu Lin, Robert Wright, Xiangyu Peng, Mustafa Hussain, Aarun Srinivas, Julia Kim, Mark O. Riedl
Researchers at Georgia Tech developed WorldCloner, a neuro-symbolic world model designed for reinforcement learning agents to quickly adjust to unforeseen environmental novelties. This approach integrates a neural policy with a rapidly updating symbolic world model, achieving up to two orders of magnitude higher policy update efficiency and robust adaptation to novelties that challenge purely neural world models.
Key insights
- A neuro-symbolic architecture effectively combines the pattern recognition strengths of neural networks with the rapid, interpretable update capabilities of symbolic reasoning to achieve robust novelty adaptation.
- Symbolic rule learning, particularly with Axis-Aligned Bounding Intervals, offers highly sample-efficient learning of environment dynamics and rapid updates in response to novelties.
- Imagination, driven by a quickly adaptable symbolic world model, can significantly reduce the real-world interaction steps and policy updates required for post-novelty policy adaptation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.