Graph Machine Learning in All Its Glory!

Our world is complex: one approach to understanding and learning more about relationships within the data is to represent it as a network or a graph - with entities as nodes and relations between them as edges. Network applications are abundant - Facebook, knowledge databases like Wikipedia, traffic routes, molecular pathways, and the fun starts when we start thinking of the physical world as graphs. Where there’s data, there’s uncertainty and the need to predict the future to make it a little less uncertain, which is why we need machine learning.

This talk will be about modeling one such network using a Graph Neural Network (GNN) and making some predictions on it using Deep Graph Library, an efficient and scalable open source framework to train and serve GNNs. While GNNs can be used for any ML task, we’ll focus on building a recommendation model.

But the transition from traditional data structures to graph-based models is not straightforward. Unlike tabular data, graphs require adapting ML principles to accommodate their topological and relational complexities.

This talk will help participants gain practical insights into modeling their data as graphs and leveraging GNNs to build a recommendation system. After the talk, the audience should have a good understanding about:
- How to structure their data as graph components and prepare it for modeling
- What a GNN is and how do we train, evaluate and serve models in DGL
- End-to-end ML tasks on graph data: best practices to avoid pitfalls