arXiv cs.AI· Akira Yoshiyama, Akmal Ashirmatov, Jonas H\"ubotter, Ido Hakimi, Andreas Krause·· 12 小时前AI 评分38
Test-Time Self-Play：无需验证器的测试时扩展方法
Test-Time Self-Play: Scaling without a Verifier
AI 导读
研究者提出 Test-Time Self-Play（TTSP），用非对称自博弈实现不依赖验证器的测试时扩展。在难度校准的 LiveCodeBench v6 子集上，TTSP 在评估的每种验证器强度下均表现最佳，且验证器越弱领先幅度越大；在无验证器极限下，mean@8 绝对提升 35%。该增益在多种验证器强度下成立，但当基础策略对正确答案覆盖接近零时除外。
正文 · 原文
Abstract:Test-time scaling methods spend additional compute at deployment to improve a large language model's performance on a hard target problem. Most methods rely on a verifier that scores candidate solutions for correctness, using it to select among or refine the model's outputs. In many deployment regimes, however, a strong verifier on the target problem is unavailable. We introduce Test-Time Self-Play (TTSP), an approach to test-time scaling using asymmetric self-play without reliance on a verifier. On a difficulty-calibrated subset of LiveCodeBench v6, TTSP is the best-performing method at every verifier strength we evaluate, and the gap to the strongest baseline widens as the verifier weakens; in the no-verifier limit, TTSP yields a 35\% absolute improvement in mean@8. The gains hold across a range of verifier strengths, except when the base policy has near-zero coverage of the correct answer.
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2611.07367 [cs.LG] |
| (or arXiv:2611.07367v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2611.07367 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Akira Yoshiyama [view email]
[v1]
Thu, 5 Nov 2026 03:07:02 UTC (4,477 KB)
来源：arXiv cs.AI · arxiv.org