PathCal: State-Aware Reflection-Marker Calibration for Efficient Reasoning
The paper introduces PathCal, a novel method for calibrating reasoning paths in Large Reasoning Language Models. It highlights the importance of distinguishing between different types of reflection markers during reasoning tasks. Experiments show that PathCal improves efficiency and performance without the need for external verifiers.
- ▪PathCal is a training-free decoding controller that focuses on state-aware reflection-marker calibration.
- ▪The study reveals that different reflection markers have distinct functional roles and influence accuracy and generation length differently.
- ▪PathCal achieves a better efficiency-performance trade-off across six reasoning benchmarks.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Artificial Intelligence arXiv:2605.23074 (cs) [Submitted on 21 May 2026] Title:PathCal: State-Aware Reflection-Marker Calibration for Efficient Reasoning Authors:Lingyu Jiang, Zirui Li, Shuo Xing, Peiran Li, Tsubasa Takahashi, Dengzhe Hou, Zhengzhong Tu, Kazunori Yamada, Fangzhou Lin View a PDF of the paper titled PathCal: State-Aware Reflection-Marker Calibration for Efficient Reasoning, by Lingyu Jiang and 8 other authors View PDF HTML (experimental) Abstract:The emergence of Large Reasoning Language Models (LRMs) has paved the way for tackling complex reasoning tasks through test-time scaling by generating long-form Chain-of-Thought (CoT) trajectories during inference.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
