EOL: Transductive Few-Shot Open-Set Recognition by Enhancing Outlier Logits

TL;DR

This paper introduces EOL, a transductive inference method for open-set few-shot recognition that improves classification and outlier detection by calibrating class prototypes and leveraging unlabelled query data.

Contribution

The paper proposes the EOL method, a novel transductive approach that enhances prototype calibration and outlier detection in open-set few-shot learning.

Findings

EOL outperforms traditional methods across multiple benchmarks.

Performance improvements range from 1.3% to 6.3%.

EOL effectively handles inlier-outlier imbalance.

Abstract

In Few-Shot Learning (FSL), models are trained to recognise unseen objects from a query set, given a few labelled examples from a support set. In standard FSL, models are evaluated on query instances sampled from the same class distribution of the support set. In this work, we explore the more nuanced and practical challenge of Open-Set Few-Shot Recognition (OSFSL). Unlike standard FSL, OSFSL incorporates unknown classes into the query set, thereby requiring the model not only to classify known classes but also to identify outliers. Building on the groundwork laid by previous studies, we define a novel transductive inference technique that leverages the InfoMax principle to exploit the unlabelled query set. We called our approach the Enhanced Outlier Logit (EOL) method. EOL refines class prototype representations through model calibration, effectively balancing the inlier-outlier ratio. This calibration enhances pseudo-label accuracy for the query set and improves the optimisation objective within the transductive inference process. We provide a comprehensive empirical evaluation demonstrating that EOL consistently surpasses traditional methods, recording performance improvements ranging from approximately $+1.3%$ to $+6.3%$ across a variety of classification and outlier detection metrics and benchmarks, even in the presence of inlier-outlier imbalance.