Overview

Knowledge Graphs [1] are becoming the standard for storing, retrieving, and querying structured data. In academia and industry, they are increasingly used to provide background knowledge. Over the last years, several research contributions were made which show that machine learning, especially representation learning, can be successfully applied to knowledge graphs enabling inductive inference about facts with unknown truth values.

Several of these approaches [2, 3] encode the graph structure that can be used for tasks such as link prediction, node classification, entity resolution, recommendation, dialogue systems, and many more. Although proposed graph representations can capture the complex relational patterns over multiple hops, they are still insufficient to solve more complex tasks such as relational reasoning [4,5]. For this kind of tasks, we envision a need for representations with more expressive power, which could include representation in non-Euclidean space. This starts by capturing e.g., type constrained, transitive or hierarchical relations in an embedding [16], up to learning expressive knowledge representations languages like first-order logic rules.

Furthermore, most approaches for learning representations for knowledge graphs focus on transductive settings, i.e., all entities and relations need to be seen during training, not allowing predictions for unseen elements [18,19]. For evolving graphs, approaches are required that generalize to unseen entities and relations. One avenue of research to address inductiveness is to employ multimodal approaches that compensate for missing modalities [20], and recently meta-learning approaches have successfully been applied [18].

Lately, the generalization of deep neural network models to non-Euclidean domains such as graphs and manifolds is explored [6]. They study the fundamental aspects that influence the underlying geometry of structured data for building graph representations [7, 8]. Recent advances in graph representation learning led to novel approaches such as convolutional neural networks for graphs [17, 9, 10, 11], attention-based graph network [12] etc. Most graphs here are either undirected or directed with both discrete and continuous node and edge attributes representing types of spatial or spectral data.

In this workshop, we want to see novel representation learning methods, approaches that can be applied to inductive learning and to (logical) reasoning [13, 14, 15], and works that shed insights into the expressive power, interpretability, and generalization of graph representation learning methods.

Also, we want to bring together researchers from different disciplines but united by their adoption of earlier mentioned techniques from machine learning. We invite the submission of papers on topics including, but not limited to:

• Knowledge graph representations for relational reasoning

• Inductive link prediction

• Graph neural networks for knowledge graphs

• Query embeddings

• Knowledge graph representation learning for conversational AI

• Unsupervised learning of complex graphs over graph-structured data

• Neural/Statistical Relational Learning

• Integrating learning of expressive knowledge representation and flexible reasoning

• Exploring non-Euclidean spaces for knowledge graph representations

• Inference tasks for learned knowledge graph representations that require general-purpose reasoning

• Entity alignment

• Knowledge graph representations for industrial recommendation systems

• Decision modeling in personalized medicine with knowledge graph representations (e.g., decision support at the point of care in tumor boards)

• Visual scene graph modeling with the help of knowledge graphs

• Knowledge graph representation to support natural language understanding

• Knowledge Graphs for cognitive science

• Representation learning on time-dependent knowledge graphs

• Question answering and commonsense reasoning via knowledge graphs

• Knowledge graph representation learning models based on adversarial methods

• Quantum Computing as a basis for scalable Knowledge graph representation learning

References

1. Färber, Michael, Frederic Bartscherer, Carsten Menne, and Achim Rettinger. "Linked data quality of dbpedia, freebase, opencyc, wikidata, and yago." Semantic Web 9, no. 1 (2018): 77-129.

2. Nickel, Maximilian, Volker Tresp, and Hans-Peter Kriegel. "A Three-Way Model for Collective Learning on Multi-Relational Data." ICML. Vol. 11. 2011.

3. Bordes, Antoine, et al. "Translating embeddings for modeling multi-relational data." Advances in neural information processing systems. 2013.

4. Ren, H., & Leskovec, J. (2020). Beta Embeddings for Multi-Hop Logical Reasoning in Knowledge Graphs. arXiv preprint arXiv:2010.11465.

5. Halford, Graeme S., William H. Wilson, and Steven Phillips. "Relational knowledge: The foundation of higher cognition." Trends in cognitive sciences 14.11 (2010): 497-505.

6. Bronstein, Michael M., et al. "Geometric deep learning: going beyond euclidean data." IEEE Signal Processing Magazine 34.4 (2017): 18-42.

7. Hamilton, William L., Rex Ying, and Jure Leskovec. "Representation learning on graphs: Methods and applications." arXiv preprint arXiv:1709.05584 (2017).

8. Nickel, Maximilian, and Douwe Kiela. "Learning continuous hierarchies in the lorentz model of hyperbolic geometry." arXiv preprint arXiv:1806.03417 (2018).

9. Niepert, Mathias, Mohamed Ahmed, and Konstantin Kutzkov. "Learning convolutional neural networks for graphs." International conference on machine learning. 2016.

10. Kipf, Thomas N., and Max Welling. "Semi-supervised classification with graph convolutional networks." arXiv preprint arXiv:1609.02907 (2016).

11. Chen, Jie, Tengfei Ma, and Cao Xiao. "Fastgcn: fast learning with graph convolutional networks via importance sampling." arXiv preprint arXiv:1801.10247 (2018).

12. Veličković, Petar, et al. "Graph attention networks." arXiv preprint arXiv:1710.10903 (2017).

13. Santoro, Adam, et al. "A simple neural network module for relational reasoning." Advances in neural information processing systems. 2017.

14. Garcia-Duran, Alberto, and Mathias Niepert. "Kblrn: End-to-end learning of knowledge base representations with latent, relational, and numerical features." arXiv preprint arXiv:1709.04676 (2017).

15. Kipf, Thomas, et al. "Neural relational inference for interacting systems." arXiv preprint arXiv:1802.04687 (2018).

16. Xie, R., Liu, Z., & Sun, M. (2016, July). Representation Learning of Knowledge Graphs with Hierarchical Types. In IJCAI (pp. 2965-2971).

17. Defferrard, M., Bresson, X., & Vandergheynst, P. (2016). Convolutional neural networks on graphs with fast localized spectral filtering. In Advances in neural information processing systems (pp. 3844-3852).

18. Baek, J., Lee, D. B., & Hwang, S. J. (2020). Learning to Extrapolate Knowledge: Transductive Few-shot Out-of-Graph Link Prediction. Advances in Neural Information Processing Systems, 33.

19. Teru, K., Denis, E., & Hamilton, W. (2020, November). Inductive relation prediction by subgraph reasoning. In International Conference on Machine Learning (pp. 9448-9457). PMLR.

20. Daza, D., Cochez, M., & Groth, P. (2020). Inductive Entity Representations from Text via Link Prediction. arXiv preprint arXiv:2010.03496.

21. Ali, M., Berrendorf, M., Hoyt, C. T., Vermue, L., Sharifzadeh, S., Tresp, V., & Lehmann, J. (2020). Pykeen 1.0: A python library for training and evaluating knowledge graph emebddings. arXiv preprint arXiv:2007.14175.