Improving Zero Shot Learning Baselines with Commonsense Knowledge
This work provides an incremental improvement for researchers and practitioners working on zero-shot learning by enhancing knowledge transfer.
This paper addresses the challenge of knowledge transfer in zero-shot learning by leveraging explicit relations from ConceptNet, a commonsense knowledge graph. They generate commonsense embeddings using a graph convolution network-based autoencoder, which, when fused with existing semantic embeddings (human-defined attributes and distributed word embeddings), surpasses strong baselines on three standard benchmark datasets.
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.