Searching Meta Reasoning Skeleton to Guide LLM Reasoning
This work addresses the limitation of inflexible meta reasoning skeletons in LLMs, offering an automated approach to improve reasoning adaptability, though it is incremental as it builds on prior skeleton-based methods.
The paper tackles the problem of manually designed meta reasoning skeletons in LLMs by proposing AutoMR, a framework that automatically searches for query-aware skeletons using a DAG representation and dynamic sampling, achieving better reasoning performance on benchmark datasets.
Meta reasoning behaviors work as a skeleton to guide large language model (LLM) reasoning, thus help to improve reasoning performance. However, prior researches implement meta reasoning skeleton with manually designed structure, limiting ability to adapt to query-specific requirement and capture intricate logical dependency among reasoning steps. To deal with the challenges, we represent meta reasoning skeleton with directed acyclic graph (DAG) to unify skeletons proposed in prior works and model intricate logical dependency. Then we propose AutoMR, a framework that searches for query-aware meta reasoning skeleton automatically inspired by automated machine learning (AutoML). Specifically, we construct search space based on DAG representation of skeleton and then formulate the search problem. We design a dynamic skeleton sampling algorithm by expanding meta reasoning skeleton along with reasoning context at inference time. This algorithm can derive any meta reasoning skeleton in search space efficiently and adapt skeleton to evolving base reasoning context, thus enable efficient query-aware skeleton search. We conduct experiments on extensive benchmark datasets. Experimental results show that AutoMR achieves better reasoning performance than previous works broadly.