Finite Automata Extraction: Low-data World Model Learning as Programs from Gameplay Video
📰 ArXiv cs.AI
Learn to extract finite automata from gameplay videos for low-data world model learning, enabling explainable and transferable environment dynamics
Action Steps
- Extract gameplay video features using computer vision techniques
- Learn a neuro-symbolic world model using Finite Automata Extraction (FAE)
- Represent the learned model as programs in a domain-specific language (DSL)
- Test and evaluate the extracted finite automata for accuracy and explainability
- Apply the learned world model to new environments for transfer learning
Who Needs to Know This
Researchers and engineers working on world model learning, neuro-symbolic AI, and explainable AI can benefit from this approach to improve environment dynamics understanding and transferability
Key Insight
💡 Finite Automata Extraction (FAE) enables learning of explainable and transferable world models from gameplay videos
Share This
🤖 Extract finite automata from gameplay videos for low-data world model learning! 📹💡
Key Takeaways
Learn to extract finite automata from gameplay videos for low-data world model learning, enabling explainable and transferable environment dynamics
Full Article
Title: Finite Automata Extraction: Low-data World Model Learning as Programs from Gameplay Video
Abstract:
arXiv:2508.11836v2 Announce Type: replace Abstract: World models are defined as a compressed spatial and temporal learned representation of an environment. The learned representation is typically a neural network, making transfer of the learned environment dynamics and explainability a challenge. In this paper, we propose an approach, Finite Automata Extraction (FAE), that learns a neuro-symbolic world model from gameplay video represented as programs in a novel domain-specific language (DSL): R
Abstract:
arXiv:2508.11836v2 Announce Type: replace Abstract: World models are defined as a compressed spatial and temporal learned representation of an environment. The learned representation is typically a neural network, making transfer of the learned environment dynamics and explainability a challenge. In this paper, we propose an approach, Finite Automata Extraction (FAE), that learns a neuro-symbolic world model from gameplay video represented as programs in a novel domain-specific language (DSL): R
DeepCamp AI