Improving Zero Shot Learning Baselines with Commonsense Knowledge

TL;DR

This paper enhances zero shot learning by integrating commonsense knowledge from ConceptNet into semantic embeddings, leading to improved transfer and classification performance on benchmark datasets.

Contribution

It introduces a method to generate commonsense embeddings using a graph convolution network autoencoder and demonstrates their effectiveness when fused with traditional semantic embeddings.

Findings

Surpassed strong baselines on three benchmark datasets.

Fusion of commonsense and semantic embeddings improves zero shot learning.

Commonsense embeddings enhance knowledge transfer in zero shot tasks.

Abstract

Zero shot learning -- the problem of training and testing on a completely disjoint set of classes -- relies greatly on its ability to transfer knowledge from train classes to test classes. Traditionally semantic embeddings consisting of human defined attributes (HA) or distributed word embeddings (DWE) are used to facilitate this transfer by improving the association between visual and semantic embeddings. In this paper, we take advantage of explicit relations between nodes defined in ConceptNet, a commonsense knowledge graph, to generate commonsense embeddings of the class labels by using a graph convolution network-based autoencoder. Our experiments performed on three standard benchmark datasets surpass the strong baselines when we fuse our commonsense embeddings with existing semantic embeddings i.e. HA and DWE.