KGTN-ens: Few-Shot Image Classification with Knowledge Graph Ensembles

TL;DR

KGTN-ens is a novel framework that enhances few-shot image classification by integrating multiple knowledge graph embeddings, including new Wikidata embeddings, leading to improved accuracy over previous methods.

Contribution

It introduces KGTN-ens, a method that combines multiple knowledge graph embeddings for better few-shot learning performance, and incorporates Wikidata embeddings as a new knowledge source.

Findings

Outperforms KGTN in top-5 accuracy on ImageNet-FS

Effective integration of multiple knowledge graph embeddings

Wikidata embeddings improve classification results

Abstract

We propose KGTN-ens, a framework extending the recent Knowledge Graph Transfer Network (KGTN) in order to incorporate multiple knowledge graph embeddings at a small cost. We evaluate it with different combinations of embeddings in a few-shot image classification task. We also construct a new knowledge source - Wikidata embeddings - and evaluate it with KGTN and KGTN-ens. Our approach outperforms KGTN in terms of the top-5 accuracy on the ImageNet-FS dataset for the majority of tested settings.