The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence) Graph ML Research at Twitter with Michael Bronstein - #394
Jul 23, 2020
Michael Bronstein, a Professor at Imperial College London and Head of Graph Machine Learning at Twitter, shares insights into the growth of graph neural networks in machine learning. He discusses challenges like scalability and dynamic graphs, delving into innovative approaches that enhance model training and expressiveness. The conversation highlights real-world applications of graph ML in drug discovery and the importance of understanding molecular properties for advancing therapies. Michael's expertise illuminates the intersection of geometry and deep learning.
AI Snips
Chapters
Transcript
Episode notes
GNNs' Rise in ML
- Graph Neural Networks (GNNs) are becoming increasingly popular in machine learning, applicable to various fields.
- Their rise is driven by larger datasets, benchmarks, improved hardware and software, similar to the deep learning boom.
Scalability Challenges for GNNs
- GNN research has focused on smaller graphs, but real-world applications often involve massive graphs.
- Scalability is a key challenge, with some companies using GNNs in production, but widespread adoption is still limited.
GNN Applicability Beyond Social Networks
- Social networks are obvious use cases for GNNs due to their inherent graph structure.
- However, GNNs are equally powerful in non-social network domains such as healthcare and physics, modeling interactions.