Rethinking Knowledge Graph Propagation for Zero-Shot Learning

By Michael C. Kampffmeyer, Yinbo Chen, Xiaodan Liang, Hao Wang, Yujia Zhang, Eric P. Xing. 2018

340 citations recorded by the source corpus.

Graph convolutional neural networks have recently shown great potential for the task of zero-shot learning. These models are highly sample efficient as related concepts in the graph structure share statistical strength allowing generalization to new classes when faced with a lack of data. However, multi-layer architectures, which are required to propagate knowledge to distant nodes in the graph, dilute the knowledge by performing extensive Laplacian smoothing at each layer and thereby consequently decrease performance. In order to still enjoy the benefit brought by the graph structure while preventing dilution of knowledge from distant nodes, we propose a Dense Graph Propagation (DGP) module with carefully designed direct links among distant nodes. DGP allows us to exploit the hierarchical graph structure of the knowledge graph through additional connections. These connections are added based on a node's relationship to its ancestors and descendants. A weighting scheme is further used to weigh their contribution depending on the distance to the node to improve information propagation in the graph. Combined with finetuning of the representations in a two-stage training approach our method outperforms state-of-the-art zero-shot learning approaches.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

Citations in the corpus

Simplifying Graph Convolutional Networks (2019, background): “ADGPM (Kampffmeyer et al., 2018) and EXEM 1-nns (Changpinyo et al.”
Label Efficient Semi-Supervised Learning via Graph Filtering (2019, methodology): “This also explains why DGPM [29] (that expands the node neighborhood by adding distant nodes) does not perform very well.”
Implicit Graph Neural Networks (2020, background): “, 2019), computer vision (Kampffmeyer et al., 2018), and recommender systems (Ying et al.”
Graph Neural Networks: A Review of Methods and Applications (2018, methodology): “ADGPM [29] uses two kinds of weight matrix, Wp and Wc, to incorporate more precise structural information.”
Large-Scale Few-Shot Learning: Knowledge Transfer With Class Hierarchy (2019, neutral): “Meta-learning based approaches [17, 21, 28, 10] have dominated.”
Region Graph Embedding Network for Zero-Shot Learning (2020, methodology): “3) It is entirely diﬀerent from GCN with word embeddings as inputs [45, 20], which learns visual classiﬁer for each class for ZSL.”
Classifier and Exemplar Synthesis for Zero-Shot Learning (2018, neutral): “Recent studies, GCNZ [74] and ADGPM [30], obtained very strong ZSL results on ImageNet .”
Generating Classification Weights With GNN Denoising Autoencoders for Few-Shot Learning (2019, methodology): “Finally, Graph Neural Networks have been applied on a different but related problem, that of zero-shot learning [40, 17], for regressing classification weights.”
Knowledge Graph Alignment Network with Gated Multi-hop Neighborhood Aggregation (2019, background): “(Kampffmeyer et al. 2019) captures the hierarchical structures of entities by introducing hypernym-hyponym links between nodes and their ancestors/descendants.”
Data-Efficient Language-Supervised Zero-Shot Learning with Self-Distillation (2021, methodology): “GCNZ[40] and DGPZ[22] use a GCN to propagate knowledge into clas-siﬁers of unseen classes, while using CNN and word embeddings to encode image and label features.”