The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence) Trends in Graph Machine Learning with Michael Bronstein - #446
Jan 11, 2021
In this engaging discussion, Michael Bronstein, a professor at Imperial College London and Head of Graph Machine Learning at Twitter, dives into the transformative power of graph machine learning. He shares insights on its applications in diverse fields like physics and bioinformatics, especially in predicting chemical properties for drug discovery. The conversation also touches on ethical considerations in AI, advancements in protein structure prediction, and the exciting future of graph ML in molecule discovery and even translating non-human communications. A must-listen for AI enthusiasts!
AI Snips
Chapters
Transcript
Episode notes
GPT-3 Progress and Limitations
- GPT-3 synthesizes human-like text, making it a potential step towards general AI.
- However, it lacks true language understanding and relies on plausible outputs.
Implicit Neural Representations
- Implicit neural representations offer a continuous, resolution-independent way to represent signals like images.
- This allows for infinite resolution and a memory footprint independent of resolution.
Halicin Discovery
- A graph neural network predicted Halicin, an antidiabetic drug, to be a powerful antibiotic.
- This demonstrates the potential of GNNs in drug discovery and repositioning.