The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence) The Physics of Data with Alpha Lee - #377
8 snips
May 21, 2020 
Alpha Lee, a Winton Advanced Fellow at Cambridge and co-founder of PostEra, delves into the fascinating intersection of physics, chemistry, and machine learning. He discusses how his startup innovates drug discovery through advanced algorithms and Bayesian approaches to manage uncertainty. The conversation highlights the parallels between physical systems and deep learning, and how transformers are revolutionizing chemical predictions. Lee also touches on recent efforts in developing effective COVID-19 therapies, underscoring the importance of collaboration in science.
AI Snips
Chapters
Transcript
Episode notes
Physics of Data
- The "physics of data" involves understanding noise and uncertainty in datasets.
- This is crucial for efficient experimentation in chemistry and material science, where experiments are costly.
Drugs and Materials
- Drug discovery and material science, while targeting different properties, share similar modeling approaches.
- Graph neural networks, particularly Bayesian ones, prove useful in both fields due to their ability to represent structures as graphs.
Physics and Loss Functions
- Alpha Lee discusses mapping machine learning loss functions to energy landscapes in physics.
- He gives examples like glasses and proteins, but hasn't yet built faster-converging architectures based on these ideas.