very cool, I was thinking about whether this kind of combo makes sense and it looks like it does! Nice work! I've also been thinking about whether Stochastic Importance Sampling or Marginal Importance Sampling could further help here somehow, perhaps for the choice of when or how often to mutate a path.
very cool, I was thinking about whether this kind of combo makes sense and it looks like it does! Nice work!
I've also been thinking about whether Stochastic Importance Sampling or Marginal Importance Sampling could further help here somehow, perhaps for the choice of when or how often to mutate a path.
In fact there already seems to be work to some of that end maybe, under the name "Conditional Resampled Importance Sampling and ReSTIR"