| layout | title | permalink |
|---|---|---|
post |
Overview |
/overview/ |
Date and time: Monday 6 May 2019, 9AM -- 6PM
Location: Room R7, Ernest N. Morial Convention Center, New Orleans
Many scientific fields study data with an underlying graph or manifold structure—such as social networks, sensor networks, biomedical knowledge graphs, and meshed surfaces in computer graphics. The need for new optimization methods and neural network architectures that can accommodate these relational and non-Euclidean structures is becoming increasingly clear. In parallel, there is a growing interest in how we can leverage insights from these domains to incorporate new kinds of relational and non-Euclidean inductive biases into deep learning.
Recent years have seen a surge in research on these problems—often under the umbrella terms of graph representation learning and geometric deep learning. For instance, new neural network architectures for graph-structured data (i.e., graph neural networks) have led to state-of-the-art results in numerous tasks—ranging from molecule classification to recommender systems—while advancements in embedding data in Riemannian manifolds (e.g., Poincaré embeddings, Hyperspherical-VAEs) and optimization on Riemannian manifolds (e.g., R-SGD, R-SVRG) have demonstrated how non-Euclidean geometries can provide powerful new kinds of inductive biases.
Perhaps the biggest testament to the increasing popularity of this area is the fact that five popular review papers have recently been published on the topic [1-5]—each attempting to unify different formulations of similar ideas across fields. This suggests that the topic has reached critical mass and requires a focused workshop to bring together researchers to identify impactful areas of interest, discuss how we can design new and better benchmarks, encourage discussion, and foster collaboration.
The workshop will consist of contributed talks, contributed posters, and invited talks on a wide variety of methods and problems in this area, including but not limited to:
- Deep learning on graphs and manifolds (e.g., graph neural networks)
- Riemannian optimization methods
- Interaction and relational networks
- Unsupervised geometric/graph embedding methods (e.g., hyperbolic embeddings)
- Generative models with manifold-valued latent variables
- Deep generative models of graphs
- Deep learning for chemical/drug design
- Deep learning on manifolds, point clouds, and for 3D vision
- Relational inductive biases (e.g., for reinforcement learning)
- Optimization challenges due to the inherent discreteness of graphs
- Theoretical analyses of graph-based and non-Euclidean machine learning approaches
- Benchmark datasets and evaluation methods
We welcome and encourage position papers under this workshop theme. We are also particularly interested in papers that introduce benchmark datasets, challenges, and competitions to further progress of the field, and we will discuss the challenge of designing such a benchmark in an interactive panel discussion.
[1] Bronstein, M. M., Bruna, J., LeCun, Y., Szlam, A., & Vandergheynst, P. (2017). Geometric deep learning: going beyond euclidean data. IEEE Signal Processing Magazine, 34(4), 18-42.
[2] Hamilton, W. L., Ying, R., & Leskovec, J. (2017). Representation learning on graphs: Methods and applications. IEEE Data Engineering Bulletin.
[3] Battaglia, P. W., Hamrick, J. B., Bapst, V., Sanchez-Gonzalez, A., Zambaldi, V., Malinowski, M., ... & Gulcehre, C. (2018). Relational inductive biases, deep learning, and graph networks. arXiv preprint arXiv:1806.01261.
[4] Goyal, P., & Ferrara, E. (2018). Graph embedding techniques, applications, and performance: A survey. Knowledge-Based Systems, 151, 78-94.
[5] Nickel, M., Murphy, K., Tresp, V., Gabrilovich, E. (2016). A review of relational machine learning for knowledge graphs. Proceedings of the IEEE. 104.1, 11-33.