Self-Supervised Approach to Addressing Zero-Shot Learning Problem

TL;DR

This paper presents a self-supervised Siamese neural network approach for zero-shot learning in entomology, achieving significant improvements in distinguishing nearly indistinguishable species.

Contribution

It introduces a novel application of contrastive learning with Siamese networks to zero-shot species identification in entomology.

Findings

61% F1-score on zero-shot instances

11% improvement on classes sharing training set intersections

Effective differentiation of similar species

Abstract

In recent years, self-supervised learning has had significant success in applications involving computer vision and natural language processing. The type of pretext task is important to this boost in performance. One common pretext task is the measure of similarity and dissimilarity between pairs of images. In this scenario, the two images that make up the negative pair are visibly different to humans. However, in entomology, species are nearly indistinguishable and thus hard to differentiate. In this study, we explored the performance of a Siamese neural network using contrastive loss by learning to push apart embeddings of bumblebee species pair that are dissimilar, and pull together similar embeddings. Our experimental results show a 61% F1-score on zero-shot instances, a performance showing 11% improvement on samples of classes that share intersections with the training set.