Academic

DenoiseFlow: Uncertainty-Aware Denoising for Reliable LLM Agentic Workflows

arXiv:2603.00532v1 Announce Type: new Abstract: Autonomous agents are increasingly entrusted with complex, long-horizon tasks, ranging from mathematical reasoning to software generation. While agentic workflows facilitate these tasks by decomposing them into multi-step reasoning chains, reliability degrades significantly as the sequence lengthens. Specifically, minor interpretation errors in natural-language instructions tend to compound silently across steps. We term this failure mode accumulated semantic ambiguity. Existing approaches to mitigate this often lack runtime adaptivity, relying instead on static exploration budgets, reactive error recovery, or single-path execution that ignores uncertainty entirely. We formalize the multi-step reasoning process as a Noisy MDP and propose DenoiseFlow, a closed-loop framework that performs progressive denoising through three coordinated stages: (1)Sensing estimates per-step semantic uncertainty; (2)Regulating adaptively allocates computati

Y
Yandong Yan, Junwei Peng, Shijie Li, Chenxi Li, Yifei Shang, Can Deng, Ruiting Dai, Yongqiang Zhao, Jiaqi Zhu, Yu Huang · · 1 min read · 3 views
#cs.AI

arXiv:2603.00532v1 Announce Type: new Abstract: Autonomous agents are increasingly entrusted with complex, long-horizon tasks, ranging from mathematical reasoning to software generation. While agentic workflows facilitate these tasks by decomposing them into multi-step reasoning chains, reliability degrades significantly as the sequence lengthens. Specifically, minor interpretation errors in natural-language instructions tend to compound silently across steps. We term this failure mode accumulated semantic ambiguity. Existing approaches to mitigate this often lack runtime adaptivity, relying instead on static exploration budgets, reactive error recovery, or single-path execution that ignores uncertainty entirely. We formalize the multi-step reasoning process as a Noisy MDP and propose DenoiseFlow, a closed-loop framework that performs progressive denoising through three coordinated stages: (1)Sensing estimates per-step semantic uncertainty; (2)Regulating adaptively allocates computation by routing between fast single-path execution and parallel exploration based on estimated risk; and (3)Correcting performs targeted recovery via influence-based root-cause localization. Online self-calibration continuously aligns decision boundaries with verifier feedback, requiring no ground-truth labels. Experiments on six benchmarks spanning mathematical reasoning, code generation, and multi-hop QA show that DenoiseFlow achieves the highest accuracy on every benchmark (83.3% average, +1.3% over the strongest baseline) while reducing cost by 40--56% through adaptive branching. Detailed ablation studies further confirm framework-level's robustness and generality. Code is available at https://anonymous.4open.science/r/DenoiseFlow-21D3/.

Executive Summary

This article proposes DenoiseFlow, a novel framework for uncertainty-aware denoising in large language model (LLM) agentic workflows. By formalizing the multi-step reasoning process as a Noisy MDP, DenoiseFlow enables progressive denoising through sensing, regulating, and correcting stages. The framework's adaptive branching and online self-calibration capabilities improve accuracy and reduce costs. Experimental results on six benchmarks demonstrate DenoiseFlow's superiority, achieving 83.3% average accuracy and reducing costs by 40-56% compared to existing approaches. Ablation studies confirm the framework's robustness and generality. DenoiseFlow's potential applications span various domains, including mathematical reasoning, code generation, and multi-hop QA.

Key Points

  • ▸ DenoiseFlow formalizes the multi-step reasoning process as a Noisy MDP
  • ▸ The framework employs progressive denoising through sensing, regulating, and correcting stages
  • ▸ Adaptive branching and online self-calibration improve accuracy and reduce costs

Merits

Strength in Formalization

DenoiseFlow's formalization of the multi-step reasoning process as a Noisy MDP provides a rigorous foundation for uncertainty-aware denoising.

Demerits

Limited Generalizability

The framework's performance on six specific benchmarks may not generalize to other domains or tasks.

Expert Commentary

DenoiseFlow represents a significant advancement in the field of uncertainty-aware LLMs. By leveraging a Noisy MDP formalization and adaptive branching, the framework addresses a critical limitation of existing approaches. However, the generalizability of DenoiseFlow's performance across different domains and tasks remains uncertain. Future research should focus on exploring the framework's applicability in novel contexts and evaluating its robustness against various types of uncertainty. Ultimately, DenoiseFlow has the potential to improve the reliability and efficiency of LLMs in a wide range of applications.

Recommendations

  • ✓ Future research should investigate DenoiseFlow's performance on diverse tasks and domains
  • ✓ Developments of DenoiseFlow should prioritize its adaptability to various types of uncertainty

Sources

Related Articles