|
|
|
|
|
|
by credit_guy
2210 days ago
|
|
|
> algorithms for plasma control and diagnostics which take advantage of new hardware (GPUs) I don't know about that. Here's a few bullet points from [1] (which someone else linked to in this thread; it describes the approach taken by these guys at TAE) that don't inspire a whole lot of confidence: "Our prior is “reasonable”, but is it really the marginal distribution over all possible plasmas? hahahahhahahaha. We model many effects, but plasmas are complex beasts and we do not model all. We only have one measurement, of much smaller dimension than our unknowns. We never sample from the tails. takes too long to get samples. by definition you can’t really validate them. Will we ever know we’re right about anything? we have zero golden data" [1] http://hyperion.usc.edu/UQ-SummerSchool/pres/Langmore.pdf |
|
|