Mixup-based Deep Metric Learning Approaches for Incomplete Supervision

TL;DR

This paper introduces three Mixup-based deep metric learning methods designed for incomplete supervision, addressing limitations of existing approaches and demonstrating superior performance across various datasets.

Contribution

The paper proposes novel Mixup-enhanced deep metric learning approaches tailored for incomplete supervision scenarios, improving robustness and effectiveness over existing methods.

Findings

Proposed approaches outperform state-of-the-art methods in multiple datasets.

Some existing metric learning methods perform poorly with incomplete supervision.

Mixup integration enhances robustness against adversarial examples.

Abstract

Deep learning architectures have achieved promising results in different areas (e.g., medicine, agriculture, and security). However, using those powerful techniques in many real applications becomes challenging due to the large labeled collections required during training. Several works have pursued solutions to overcome it by proposing strategies that can learn more for less, e.g., weakly and semi-supervised learning approaches. As these approaches do not usually address memorization and sensitivity to adversarial examples, this paper presents three deep metric learning approaches combined with Mixup for incomplete-supervision scenarios. We show that some state-of-the-art approaches in metric learning might not work well in such scenarios. Moreover, the proposed approaches outperform most of them in different datasets.