KEYNOTE: Python and the Future of Fusion Energy 🌳
summary
Talk claims breakeven at the National Ignition Facility and urges software help for fusion. Concrete work combines multiple diagnostics with models into a unified estimate of plasma temperature, uses Markov chain Monte Carlo in Python with numpy and scipy, speeds up with precomputed tables and struggles with cross-language serialization. A planned device called eater generates roughly 2 petabytes per day and needs unified access. Machine learning predicts disruptions with results like 90% true positive at 5% false positive but fails to extrapolate across machine scales.
AI-written summary · about
chapters (13)
- 0:00Intro
- 2:15Fusion is a reaction where nuclei of two light atoms bo together and release energy
- 4:37Fusion energy promises safe, clean, abundant electricity
- 6:39Power plant grade plasmas must be hot and dense
- 9:22Magnetic fields confine plasmas to protect walls from melting
- 10:54Magnetically confined plasmas are highly dynamic systems
- 13:12Data aggregation can address the limitations of the harsh fusion environment
- 14:47Python is an excellent tool for this
- 17:45Python is a standard tool for scientific computing in fusion
- 21:33ITER Integrated Modelling & Analysis Suite
- 25:12Higher power plasmas can become catastrophically unstable
- 31:02ML is increasingly being used as a tool for feedback and control of fusion devices
- 32:37Fusion power is a growing industry
related talks
23mKeynote: AI, Quantum, and AI + Quantum
21mKeynote: A Tale of Using Kubeflow to Make the Electricity Smarter in China
20mKeynote: Noah Goodman, Uber AI Labs, Pyro, Stanford University in conversation with Jim Zemlin
16mGame-changing Engineering
37mKeynote – Carol Willing
24mFIRESIDE CHAT: The Future of AI Infra: The Opportunities and Challenges That Await Us On Our Journey
06mThe AI Compute Crisis & Web3 Opportunity
58mThe Next Decade of Software Development