Spaces:

---

calibrationcomp

/

calibration_benchmark

Like 0

Running

score-based methods significantly improve the base derivative methods,

but still suffer from "how" one computes the score - KGMM works efficiently compared to denoising score matching (DSM), in low dimensions but both have efficiency issues in higher dimensions. I have already iterated several times to obtain improvements with steadily more technical robustness guardrails. Will post these preliminaries now, but neaten up later