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

2025-02-26

AI Summary

To improve LLM prediction performance during inference, this paper proposes TestNUC—a test-time adaptation method that dynamically refines predictions by leveraging local consistency among unlabeled samples in the input’s neighborhood. Its core contribution is the first linearly scalable test-time computation paradigm, integrating neighborhood consistency modeling, unsupervised similarity retrieval, and embedding-space proximity analysis. Crucially, TestNUC seamlessly interoperates with existing test-time methods (e.g., self-consistency). Evaluated across eight diverse benchmark datasets spanning multiple NLP tasks, TestNUC consistently outperforms standard prompting, self-consistency, and other baselines. Prediction accuracy improves steadily with increasing numbers of neighborhood unlabeled samples, demonstrating robust scalability. Moreover, TestNUC is model-agnostic with respect to embedding architectures, maintaining compatibility across various off-the-shelf embedding models. This work advances test-time learning by unifying consistency-driven refinement with efficient, plug-and-play neighborhood exploitation.

Technology Category

• Natural Language Processing: Sentence-level Semantics, Textual Inference, etc.
• Machine Learning: Large Multimodal Models (LMMs)
• Search and Optimization: Learning to Search

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

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.

Problem

• Enhances test-time predictions
• Leverages unlabeled data consistency
• Improves large language model performance

Innovation

• Leverages neighboring unlabeled data consistency
• Linearly scales with unlabeled data
• Seamlessly integrates with existing test-time approaches