TestNUC: Enhancing Test-Time Computing Approaches and Scaling through Neighboring Unlabeled Data Consistency

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Zou, Henry Peng, Gu, Zhengyao, Zhou, Yue, Chen, Yankai, Zhang, Weizhi, Fang, Liancheng, Wang, Yibo, Li, Yangning, Liu, Kay, Yu, Philip S.
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916770075377664
author Zou, Henry Peng
Gu, Zhengyao
Zhou, Yue
Chen, Yankai
Zhang, Weizhi
Fang, Liancheng
Wang, Yibo
Li, Yangning
Liu, Kay
Yu, Philip S.
author_facet Zou, Henry Peng
Gu, Zhengyao
Zhou, Yue
Chen, Yankai
Zhang, Weizhi
Fang, Liancheng
Wang, Yibo
Li, Yangning
Liu, Kay
Yu, Philip S.
contents Test-time computing approaches, which leverage additional computational resources during inference, have been proven effective in enhancing large language model performance. This work introduces a novel, linearly scaling approach, TestNUC, that improves test-time predictions by leveraging the local consistency of neighboring unlabeled data-it classifies an input instance by considering not only the model's prediction on that instance but also on neighboring unlabeled instances. We evaluate TestNUC across eight diverse datasets, spanning intent classification, topic mining, domain discovery, and emotion detection, demonstrating its consistent superiority over baseline methods such as standard prompting and self-consistency. Furthermore, TestNUC can be seamlessly integrated with existing test-time computing approaches, substantially boosting their performance. Our analysis reveals that TestNUC scales effectively with increasing amounts of unlabeled data and performs robustly across different embedding models, making it practical for real-world applications. Our code is available at https://github.com/HenryPengZou/TestNUC.
format Preprint
id arxiv_https___arxiv_org_abs_2502_19163
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle TestNUC: Enhancing Test-Time Computing Approaches and Scaling through Neighboring Unlabeled Data Consistency
Zou, Henry Peng
Gu, Zhengyao
Zhou, Yue
Chen, Yankai
Zhang, Weizhi
Fang, Liancheng
Wang, Yibo
Li, Yangning
Liu, Kay
Yu, Philip S.
Computation and Language
Artificial Intelligence
Information Retrieval
Machine Learning
Test-time computing approaches, which leverage additional computational resources during inference, have been proven effective in enhancing large language model performance. This work introduces a novel, linearly scaling approach, TestNUC, that improves test-time predictions by leveraging the local consistency of neighboring unlabeled data-it classifies an input instance by considering not only the model's prediction on that instance but also on neighboring unlabeled instances. We evaluate TestNUC across eight diverse datasets, spanning intent classification, topic mining, domain discovery, and emotion detection, demonstrating its consistent superiority over baseline methods such as standard prompting and self-consistency. Furthermore, TestNUC can be seamlessly integrated with existing test-time computing approaches, substantially boosting their performance. Our analysis reveals that TestNUC scales effectively with increasing amounts of unlabeled data and performs robustly across different embedding models, making it practical for real-world applications. Our code is available at https://github.com/HenryPengZou/TestNUC.
title TestNUC: Enhancing Test-Time Computing Approaches and Scaling through Neighboring Unlabeled Data Consistency
topic Computation and Language
Artificial Intelligence
Information Retrieval
Machine Learning
url https://arxiv.org/abs/2502.19163