analiza aiqbench
This research introduces DDRG, a framework that enhances large language model reasoning by using graph alignment to resolve intermediate claim disagreements without expanding the sampled trace bank. By directing localized conflicts to an answer-blind verifier, the method achieves significant accuracy improvements over standard self-consistency and chain-of-thought approaches on multiple benchmarks. The work represents a meaningful advancement in fine-grained correction techniques for multi-step reasoning tasks within current LLM architectures.
Beyond Answer Voting: Fine-Grained Correction of LLM Reasoning through Graph Alignment across Sampled Traces
arXiv:2610.18606v1 Announce Type: new Abstract: Large language models increasingly tackle tasks that require multi-step reasoning, where errors in intermediate claims can affect the final answer. Self-consistency (SC) improves reasoning by voting over sampled final answers, but its vote leaves intermediate disagreements unresolved. To address this problem, we introduce DDRG (Disagreement-Driven Verification Guided by Reasoning Graphs), a framework for selective correction from a frozen bank of reasoning traces. Our insight is that graph alignment can expose claim-level disagreements and provide correction evidence without expanding the sampled trace bank. DDRG builds source-grounded claim graphs and sends localized conflicts to an answer-blind verifier. A verifier-supported refutation that reaches the majority answer's supporting path triggers one focused repair; otherwise, DDRG returns the exact SC vote. Experimental results across five benchmarks show that native zero-shot DDRG improves macro-average accuracy over direct answering by 8.65 and 15.83 percentage points for Qwen2.5-7B and Llama3.1-8B. It also exceeds CoT, the evaluated GoT and GoV question-answering adaptations, and SC on this macro metric. DDRG links local verification to selective correction without enlarging the initial candidate bank.
Abstract:Large language models increasingly tackle tasks that require multi-step reasoning, where errors in intermediate claims can affect the final answer. Self-consistency (SC) improves reasoning by voting over sampled final answers, but its vote leaves intermediate disagreements unresolved. To address this problem, we introduce DDRG (Disagreement-Driven Verification Guided by Reasoning Graphs), a framework for selective correction from a frozen bank of reasoning traces. Our insight is that graph alignment can expose claim-level disagreements and provide correction evidence without expanding the sampled trace bank. DDRG builds source-grounded claim graphs and sends localized conflicts to an answer-blind verifier. A verifier-supported refutation that reaches the majority answer's supporting path triggers one focused repair; otherwise, DDRG returns the exact SC vote. Experimental results across five benchmarks show that native zero-shot DDRG improves macro-average accuracy over direct answering by 8.65 and 15.83 percentage points for Qwen2.5-7B and Llama3.1-8B. It also exceeds CoT, the evaluated GoT and GoV question-answering adaptations, and SC on this macro metric. DDRG links local verification to selective correction without enlarging the initial candidate bank.
19 pages, 5 figures, 4 tables. Under review
Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
arXiv:2610.18606 [cs.AI]
(or arXiv:2610.18606v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2610.18606 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 278 cuvinte
Subiecte conexe
Beyond Exact Match: Task-Aware GRPO for Cross-Domain PCBA Visual Question Answering
arXiv · 22 sept. 2026
OpenMedReason: Scientific Reasoning Supervision for Medical Vision-Language Models
arXiv · 12 iun. 2026
See, Measure, and Reason: Learning Visually Grounded Reasoning in Pathology
arXiv · 29 sept. 2026
SAFE-G: Structure-aware Faithful Evidence-guided Generation for Knowledge-based Visual Question Answering
arXiv · 25 aug. 2026