Closed JinraeKim closed 1 year ago
Optimization methods that form part of Surrogates require additional data point acquisition.
Yes, these methods are for surrogate optimization, which requires acquiring more data points to improve the surrogate and then choose the next point. That's kind of the point of these methods.
If you're looking for accelerating optimization by a surrogate, that's a different thing. For that, just train a surrogate of your model/loss and then put that into Optimization.jl. Since that doesn't require any new tooling, we just recommend using Optimization.jl there.
I was looking at this package and it seems that the optimization methods for surrogates such as SRBF requires additional data point acquisition. Is there any optimization method here which does not require additional data acquisition?