Out-of-Distribution Detection by Leveraging Between-Layer Transformation Smoothness

TL;DR

This paper introduces BLOOD, a novel method for out-of-distribution detection in Transformers that leverages the smoothness of transformations between layers, working effectively without access to training data.

Contribution

BLOOD is a new OOD detection approach that exploits between-layer transformation smoothness in pre-trained Transformers, applicable without training data access.

Findings

BLOOD outperforms comparable resource methods on text classification tasks.

In simpler tasks, OOD transformations remain sharp; in complex tasks, they become smoother.

Transformation smoothness correlates with task complexity.

Abstract

Effective out-of-distribution (OOD) detection is crucial for reliable machine learning models, yet most current methods are limited in practical use due to requirements like access to training data or intervention in training. We present a novel method for detecting OOD data in Transformers based on transformation smoothness between intermediate layers of a network (BLOOD), which is applicable to pre-trained models without access to training data. BLOOD utilizes the tendency of between-layer representation transformations of in-distribution (ID) data to be smoother than the corresponding transformations of OOD data, a property that we also demonstrate empirically. We evaluate BLOOD on several text classification tasks with Transformer networks and demonstrate that it outperforms methods with comparable resource requirements. Our analysis also suggests that when learning simpler tasks, OOD data transformations maintain their original sharpness, whereas sharpness increases with more complex tasks.