FUSE: Ensembling Verifiers with Zero Labeled Data

TL;DR

FUSE is a novel unsupervised ensembling method that enhances model output verification without needing labeled data, improving accuracy across various benchmarks.

Contribution

Introduces FUSE, a zero-label ensembling technique that controls verifier dependencies to boost verification performance without ground truth labels.

Findings

FUSE matches or outperforms semi-supervised methods in diverse benchmarks.

Effective across academic and frontier verification tasks.

Improves verification quality without labeled data.

Abstract

Verification of model outputs is rapidly emerging as a key primitive for both training and real-world deployment of large language models (LLMs). In practice, this often involves using imperfect LLM judges and reward models since ground truth acquisition can be time-consuming and expensive. We introduce Fully Unsupervised Score Ensembling (FUSE), a method for improving verification quality by ensembling verifiers without access to ground truth correctness labels. The key idea behind FUSE is to control conditional dependencies between verifiers in a manner that improves the unsupervised performance of a class of spectral algorithms from the ensembling literature. Despite requiring zero ground truth labels, FUSE typically matches or improves upon semi-supervised alternatives in test-time scaling experiments with diverse sets of generator models, verifiers, and benchmarks. In particular, we validate our method on both conventional academic benchmarks such as GPQA Diamond and on frontier, unsaturated benchmarks such as Humanity's Last Exam and IMO Shortlist questions.