Test-Time Scaling via Budgeted Multi-Attribute Verification

arXiv:2609.34322v2 Announce Type: replace Abstract: Verifying LLM-generated answers under a shared computational budget requires jointly deciding which candidates to inspect and which verification attributes to evaluate. We formulate this problem as multi-attribute good-arm identification under a global budget: each candidate is an arm evaluated along several costly attributes, and the goal is to certify as many candidates as possible whose mean scores exceed the prescribed thresholds on all attributes. We propose extsc{BMA-GAI}, an algorithm that combines cost-aware arm selection with adaptive sampling of attributes. Every observation serves both to guide adaptive allocation and to support anytime-valid certification, which removes the need for a separate confirmation stage. We establish an asymptotic coverage guarantee for extsc{BMA-GAI} and derive a matching information-theoretic converse that characterizes the intrinsic complexity of the problem, thereby proving that extsc{BMA-GAI} is first-order optimal away from critical budget levels. Experiments on synthetic benchmarks and an LLM answer-verification task show that extsc{BMA-GAI} allocates the verification budget more efficiently and certifies more high-quality candidates than competing methods.

Related

• Strategic Scaling of Test-Time Compute: A Bandit Learning Approach
• Reliable Self-Improvement Training by Verifying Reasoning, Not Just Answers
• ThinkRetrieve: Retrieval-Augmented Reasoning Traces for Test-Time Scaling
• Selective Test-Time Compute Scaling for Click-Through Rate Prediction via Uncertainty-Triggered Feature Path Exploration

Source: arXiv cs.AI | 2026-10-15