Huge kudos to the team for their hard work in delivering this high-quality contribution to the community. Open-source fusion modelling just got a big boost!
Paper:
pubs.aip.org/aip/pop/arti...
FreeGSNKE is the UK Fusion Computing Lab's answer to this challenge!
↔️ Fully benchmarked against existing codes
🦾 Robust solvers
🖥️ Modern coding practices
📖 Excellent documentation and tutorials
🌐 Fully open source
github.com/FusionComput...
🧵2/3
🚀FreeGSNKE is now open-source! Big milestone, esp for those interested in ML/RL for fusion🚀
I've talked to many people who have been inspired by Deepmind's 2022 Nature paper on tokamak magnetic control, but struggled to get involved in ML-control for fusion problems...
🧵1/3
Interested in symmetries/GPs/BayesOpt/nuclear fusion? Come and check out our #NeurIPS poster!
📍West Ballroom #6003 @ 11am-2pm
🔍 Not sure if it's for you? Read the blog post to get a brief insight >>> theobrown.uk/blog/invaria...
w/ @ilijabogunovic.bsky.social
@uclofficial.bsky.social
📝 In our paper, we back up our empirical findings with upper and lower regret bounds.
Check it out! arxiv.org/abs/2410.16972
#NeurIPS2024 #research #ml
🧵 4/4
⚡ Invariant BO has a whole host of applications. We tackle a difficult design task from nuclear fusion: designing a heating system for STEP, the UK's flagship next-gen fusion reactor.
🧵 3/4
#STEPtoFusion #fusionenergy
⬆️ Using an invariant kernel massively boosts sample efficiency (red line).
For a low-cost approximation, you can use a *subset* of symmetries without sacrificing improved performance (yellow + green lines).
🧵 2/4
📣 If you've got an objective that exhibits symmetries, you should be using invariant kernel BO 📣
🚀 More sample efficient than constrained/naive BO!
🚀 More compute efficient than data augmentation!
🧵 1/4
#NeurIPS2024 #BayesianOptimisation #ai
