SCOUT: Teaching Pre-trained Language Models to Enhance Reasoning via Flow Chain-of-Thought

2025-05-30

AI Summary

Existing chain-of-thought (CoT) methods rely on labor-intensive manual annotation of intermediate reasoning steps, limiting generalizability; recursive reasoning approaches often require costly pretraining and lack theoretical modeling of iterative evolution. This paper proposes Flow CoT—a paradigm that models reasoning as progressive evolution of latent cognitive states—requiring neither explicit CoT supervision nor additional pretraining. We further introduce SCOUT, a lightweight fine-tuning framework integrating progressive knowledge distillation and cross-step backtracking attention to enable multi-round deep reasoning. Key contributions include: (i) the first formalization of reasoning as cognitive trajectory evolution; (ii) the first scalable, unsupervised, pretraining-free recursive reasoning method; and (iii) a staged teacher-alignment strategy coupled with a backtracking mechanism that preserves the original computational flow. Evaluated on eight reasoning benchmarks, Flow CoT achieves an average accuracy gain of +1.8%, alongside substantial improvements in explanation quality and reasoning depth.

Technology Category

• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning
• Reasoning under Uncertainty: Sequential Decision Making
• Knowledge Representation and Reasoning: Computational Complexity of Reasoning

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Chain of Thought (CoT) prompting improves the reasoning performance of large language models (LLMs) by encouraging step by step thinking. However, CoT-based methods depend on intermediate reasoning steps, which limits scalability and generalization. Recent work explores recursive reasoning, where LLMs reuse internal layers across iterations to refine latent representations without explicit CoT supervision. While promising, these approaches often require costly pretraining and lack a principled framework for how reasoning should evolve across iterations. We address this gap by introducing Flow Chain of Thought (Flow CoT), a reasoning paradigm that models recursive inference as a progressive trajectory of latent cognitive states. Flow CoT frames each iteration as a distinct cognitive stage deepening reasoning across iterations without relying on manual supervision. To realize this, we propose SCOUT (Stepwise Cognitive Optimization Using Teachers), a lightweight fine tuning framework that enables Flow CoT style reasoning without the need for pretraining. SCOUT uses progressive distillation to align each iteration with a teacher of appropriate capacity, and a cross attention based retrospective module that integrates outputs from previous iterations while preserving the models original computation flow. Experiments across eight reasoning benchmarks show that SCOUT consistently improves both accuracy and explanation quality, achieving up to 1.8% gains under fine tuning. Qualitative analyses further reveal that SCOUT enables progressively deeper reasoning across iterations refining both belief formation and explanation granularity. These results not only validate the effectiveness of SCOUT, but also demonstrate the practical viability of Flow CoT as a scalable framework for enhancing reasoning in LLMs.

Problem

• Enhancing reasoning in LLMs without explicit CoT supervision
• Reducing dependency on costly pretraining for recursive reasoning
• Improving accuracy and explanation quality in reasoning tasks

Innovation

• Flow CoT models recursive inference as progressive cognitive states
• SCOUT enables Flow CoT reasoning without pretraining via fine-tuning
• Progressive distillation aligns iterations with teacher capacities