arXiv：cs.AI· Cong Hua, Qianqian Xu, Zitai Wang, Zhiyong Yang, Shilong Bao, Linchao Zhang, Qingming Huang·· 16 小时前
PGLA：面向 LLM 测试时扩展的两两引导 Logit 调整方法
Compare to Correct: Pairwise-Guided Logit Adjustment for LLM Test-Time Scaling
AI 导读
研究者提出 Pairwise-Guided Logit Adjustment（PGLA），借助社会选择理论中的 Condorcet 原则，对支持不同答案的推理路径做两两置信度比较，并通过 KL 正则化目标将这种比较证据与投票支持结合。
正文
Abstract:Sampling-based test-time scaling enhances LLM reasoning by aggregating multiple reasoning paths for the same query. Existing methods use answer frequencies or path confidence to compute candidate scores, which do not explicitly characterize comparisons between competing answers. Inspired by the Condorcet principle in social choice theory, we introduce pairwise confidence comparisons between paths supporting different answers to capture answer-level comparative advantage. We combine this comparative evidence with voting support through a KL-regularized objective. Theoretically, we show that this comparative-advantage is equivalent to the adjustment to the traditional voting logits. In this way, our proposed Pairwise-Guided Logit Adjustment (PGLA) framework allows comparative evidence to compensate for weaker voting support and correct voting-based ranking errors. Experiments on 10 challenging benchmarks including mathematical reasoning, knowledge reasoning, and code generation with 7 models ranging from 1.7B to 32B parameters speak to the effectiveness of this plug-and-play term. Additional evaluations with external reward scores and online sampling further demonstrate its applicability.
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.18583 [cs.AI] |
| (or arXiv:2610.18583v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.18583 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Cong Hua [view email]
[v1]
Tue, 13 Oct 2026 15:46:15 UTC (5,637 KB)
来源：arXiv：cs.AI · arxiv.org