4 ms·

This problem is clearly a fabulous use case for machine learning.

vrm 5y ago
like I said above, we certainly hope so! It has been slow progress so far but applying modern ML / control techniques to tokamaks is one of the truly exciting applications of the current generation of AI in my opinion. Biased because this is literally what I do all day
- mrDmrTmrJ 5y ago
  Do you have a website or any papers on your work I could read?
  - vrm 5y ago
    I need to redo my website, just getting into the more public part of my PhD. Papers I would recommend from our collaboration on control of normalized plasma pressure: https://papers.nips.cc/paper/2019/hash/7876acb66640bad41f1e1371ef30c180-Abstract.html https://papers.nips.cc/paper/2019/hash/7876acb66640bad41f1e1... plasma profile transport modeling: https://iopscience.iop.org/article/10.1088/1741-4326/abe08d/meta https://iopscience.iop.org/article/10.1088/1741-4326/abe08d/... hybrid dynamical modeling of gross plasma quantities: https://arxiv.org/abs/2006.12682 https://arxiv.org/abs/2006.12682 uncertainty quantification for plasma dynamics: https://arxiv.org/abs/2011.09588 https://arxiv.org/abs/2011.09588 It's still early days for this work and for us but we're looking at pushing reinforcement learning in methods and engineering to solve this problem
    - PicassoCTs 5y ago
      Do you have a github repo for the controller software? It will be fiberop to the sensors? And currentcontrolling by the ai? Have you considered antagonistic training? One AI tries to destabilize the proces, the other trains not against a simulation, but against the destabilizing input and a succes-metric?
willis936 5y ago
Yes but perhaps moreso in the more traditional ways. Stellarator geometries have several layers of optimization. You optimize for a last closed flux surface weighing MHD and turbulence models, then you optionally optimize nested flux surfaces, then you optimize single filament coils, then you optimized multi-filament coils. The computations are all too much, but more efficient optimizations buy performance for less computing.
- plutonorm 5y ago
  I don't think so. Don't prematurely optimise. First thing to try is to hook up an RL algorithm and let her rip. If that doesn't work move on from there.

vrm 5y ago

like I said above, we certainly hope so! It has been slow progress so far but applying modern ML / control techniques to tokamaks is one of the truly exciting applications of the current generation of AI in my opinion. Biased because this is literally what I do all day

mrDmrTmrJ 5y ago
Do you have a website or any papers on your work I could read?
- vrm 5y ago
  I need to redo my website, just getting into the more public part of my PhD. Papers I would recommend from our collaboration on control of normalized plasma pressure: https://papers.nips.cc/paper/2019/hash/7876acb66640bad41f1e1371ef30c180-Abstract.html https://papers.nips.cc/paper/2019/hash/7876acb66640bad41f1e1... plasma profile transport modeling: https://iopscience.iop.org/article/10.1088/1741-4326/abe08d/meta https://iopscience.iop.org/article/10.1088/1741-4326/abe08d/... hybrid dynamical modeling of gross plasma quantities: https://arxiv.org/abs/2006.12682 https://arxiv.org/abs/2006.12682 uncertainty quantification for plasma dynamics: https://arxiv.org/abs/2011.09588 https://arxiv.org/abs/2011.09588 It's still early days for this work and for us but we're looking at pushing reinforcement learning in methods and engineering to solve this problem
  - PicassoCTs 5y ago
    Do you have a github repo for the controller software? It will be fiberop to the sensors? And currentcontrolling by the ai? Have you considered antagonistic training? One AI tries to destabilize the proces, the other trains not against a simulation, but against the destabilizing input and a succes-metric?

willis936 5y ago

Yes but perhaps moreso in the more traditional ways. Stellarator geometries have several layers of optimization. You optimize for a last closed flux surface weighing MHD and turbulence models, then you optionally optimize nested flux surfaces, then you optimize single filament coils, then you optimized multi-filament coils. The computations are all too much, but more efficient optimizations buy performance for less computing.

plutonorm 5y ago
I don't think so. Don't prematurely optimise. First thing to try is to hook up an RL algorithm and let her rip. If that doesn't work move on from there.