5 ms·
Calling it "AI" is a stretch, it's more like using the TensorFlow infrastructure to do Bayesian inference. Here's a presentation on the process: http://hyperion
by oddthink 6y ago
Calling it "AI" is a stretch, it's more like using the TensorFlow infrastructure to do Bayesian inference. Here's a presentation on the process: http://hyperion.usc.edu/UQ-SummerSchool/pres/Langmore.pdf http://hyperion.usc.edu/UQ-SummerSchool/pres/Langmore.pdf
- ikeyany 6y agoSo their only paragraph on it, > Artificial intelligence software, recently developed by TAE with help from Google, should keep the fusion fuel stable even at the 3-billion-degree temperatures needed in a commercial-scale power plant. “A year ago I was concerned about the ability to control the machine,” Binderbauer says, “but no longer.”Courtesy of TAE Technologies which implies that it's being used for plasma control, is a lie.
- oddthink 6y agoNo, it's used for control. But you should be thinking more like "process control" than "intelligent" control. Or to be more specific, deriving an optimal process control model via pretty hard-core plasma density reconstructions; my understanding is that they track plasma density fluctuations and tweak parameters in real-time, but I would be surprised if the TF model is in the control path. It seems a bit overly-aggressive to accuse them of lying because some pop-sci journalist heard tensorflow and went straight to "artificial intelligence." And even there, it's an understandable and common hype. It doesn't take away from the point that they're doing some really interesting and novel computational physics to make this thing work.