Modeling Hierarchical Thinking in Large Reasoning Models
📰 ArXiv cs.AI
Learn to model hierarchical thinking in large reasoning models using finite state machines to improve consistency and reduce reasoning pathologies
Action Steps
- Define the six abstract cognitive states for hierarchical reasoning
- Implement a Finite State Machine (FSM) to model the transitions between these states
- Train the FSM using Chain-of-Thought (CoT) sequences from Large Reasoning Models (LRMs)
- Evaluate the performance of the FSM in approximating LRM's emerging hierarchical reasoning dynamics
- Refine the FSM by incorporating additional cognitive states or transitions as needed
Who Needs to Know This
AI researchers and engineers working on large reasoning models can benefit from this approach to improve the consistency and effectiveness of their models
Key Insight
💡 Modeling hierarchical thinking in LRMs as a Finite State Machine can help reduce inconsistencies and reasoning pathologies
Share This
🤖 Improve Large Reasoning Models with hierarchical thinking using Finite State Machines! 📈
Key Takeaways
Learn to model hierarchical thinking in large reasoning models using finite state machines to improve consistency and reduce reasoning pathologies
Full Article
Title: Modeling Hierarchical Thinking in Large Reasoning Models
Abstract:
arXiv:2510.22437v2 Announce Type: replace Abstract: Large Reasoning Models (LRMs) solve complex tasks by generating long Chain-of-Thought (CoT) sequences; however, the emergent dynamics governing reasoning trajectories are not well understood and can lead to inconsistencies and reasoning pathologies. In this work, we propose to approximate LRM's emerging hierarchical reasoning dynamics as a trajectory within a Finite State Machine (FSM) transitioning among six abstract cognitive states. We demon
Abstract:
arXiv:2510.22437v2 Announce Type: replace Abstract: Large Reasoning Models (LRMs) solve complex tasks by generating long Chain-of-Thought (CoT) sequences; however, the emergent dynamics governing reasoning trajectories are not well understood and can lead to inconsistencies and reasoning pathologies. In this work, we propose to approximate LRM's emerging hierarchical reasoning dynamics as a trajectory within a Finite State Machine (FSM) transitioning among six abstract cognitive states. We demon
DeepCamp AI