Fine-Tuning Deteriorates General Textual Out-of-Distribution Detection by Distorting Task-Agnostic Features

Abstract

Detecting out-of-distribution (OOD) inputs is crucial for the safe deployment of natural language processing (NLP) models. Though existing methods, especially those based on the statistics in the feature space of fine-tuned pretrained language models (PLMs), are claimed to be effective, their effectiveness on different types of distribution shifts remains underexplored. In this work, we take the first step to comprehensively evaluate the mainstream textual OOD detection methods for detecting semantic and non-semantic shifts. We find that: (1) no existing method behaves well in both settings; (2) fine-tuning PLMs on indistribution data benefits detecting semantic shifts but severely deteriorates detecting nonsemantic shifts, which can be attributed to the distortion of task-agnostic features. To alleviate the issue, we present a simple yet effective general OOD score named GNOME that integrates the confidence scores derived from the task-agnostic and task-specific representations. Experiments show that GNOME works well in both semantic and non-semantic shift scenarios, and further brings significant improvement on two cross-task benchmarks where both kinds of shifts simultaneously take place. Our code is available at https:// github.com/lancopku/GNOME.

FLatS: Principled Out-of-Distribution Detection with Feature-Based Likelihood Ratio Score

“…We compare the proposed FLatS with 9 popular OOD detection methods. (1) For confidence-based methods that leverages output probabilities of classifiers trained on IND data to detect OOD samples, we evaluate MSP , energy score (Liu et al, 2020), ODIN (Liang et al, 2017), D2U (Chen et al, 2023), MLS (Hendrycks et al, 2019); (2) For distancebased methods, we test LOF (Breunig et al, 2000), Maha , KNN , and GNOME (Chen et al, 2023).…”

“…Hyperparameters. We use k = 10 for KNN following (Chen et al, 2023). Searching from {0.1, 0.2, 0.5, 1.0, 2.0}, we adopt α = 0.5 for Equation (5).…”