Mixup-based Deep Metric Learning Approaches for Incomplete Supervision
This work addresses the challenge of requiring large labeled datasets for deep learning in real applications like medicine and security, offering an incremental improvement for incomplete-supervision scenarios.
The paper tackles the problem of deep metric learning in incomplete-supervision scenarios by proposing three approaches that combine deep metric learning with Mixup, showing that some state-of-the-art methods perform poorly in these settings and that the proposed methods outperform most of them across different datasets.
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.