KGTN-ens: Few-Shot Image Classification with Knowledge Graph Ensembles
This work addresses few-shot learning for image classification, but it is incremental as it builds upon an existing method with minor enhancements.
The paper tackles few-shot image classification by extending the Knowledge Graph Transfer Network (KGTN) to incorporate multiple knowledge graph embeddings at low cost, resulting in improved top-5 accuracy on the ImageNet-FS dataset compared to KGTN in most settings.
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.