Adapting Feature Attenuation to NLP

2026-01-02

arXiv.org

AI Summary

This work addresses the brittleness of language models in open-set recognition (OSR) when encountering unseen classes by adapting the COSTARR framework—originally developed in computer vision—to natural language processing. The approach is applied to BERT and GPT-2 without requiring retraining, thereby enhancing their OSR capabilities. A comprehensive evaluation across 176 arXiv subject classification tasks compares COSTARR against established baselines, including Maximum Softmax Probability (MSP), MaxLogit, and temperature-scaled free energy scoring. Results demonstrate that COSTARR can be directly transferred to NLP and effectively improves OSR performance; however, it does not significantly outperform MSP or MaxLogit in high-cardinality settings. Notably, free energy scoring consistently underperforms, highlighting inherent limitations in transferring vision-based OSR methods to language models.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Computer Vision: Language and Vision

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Transformer classifiers such as BERT deliver impressive closed-set accuracy, yet they remain brittle when confronted with inputs from unseen categories--a common scenario for deployed NLP systems. We investigate Open-Set Recognition (OSR) for text by porting the feature attenuation hypothesis from computer vision to transformers and by benchmarking it against state-of-the-art baselines. Concretely, we adapt the COSTARR framework--originally designed for classification in computer vision--to two modest language models (BERT (base) and GPT-2) trained to label 176 arXiv subject areas. Alongside COSTARR, we evaluate Maximum Softmax Probability (MSP), MaxLogit, and the temperature-scaled free-energy score under the OOSA and AUOSCR metrics. Our results show (i) COSTARR extends to NLP without retraining but yields no statistically significant gain over MaxLogit or MSP, and (ii) free-energy lags behind all other scores in this high-class-count setting. The study highlights both the promise and the current limitations of transplanting vision-centric OSR ideas to language models, and points toward the need for larger backbones and task-tailored attenuation strategies.

Problem

• Open-Set Recognition
• Transformer
• Feature Attenuation
• NLP
• Out-of-Distribution Detection

Innovation

• Open-Set Recognition
• Feature Attenuation
• COSTARR
• Transformer
• Out-of-Distribution Detection