LLNL / macc

Robust neural network surrogate for inertial confinement fusion
MIT License
26 stars 9 forks source link

Uncertainty Measures in ICF-CycleGAN #1

Open diliprk opened 5 years ago

diliprk commented 5 years ago

Hi Rushil, Thank you for sharing this interesting work in GitHub. I also have a requirement of using CycleGANs to synthesise simulated data (for a regression problem) matching the distribution of some lab test results and will be citing your work 👍 .

In the presentation of your ICF CycleGAN project by Brian K.Spears, I saw some Bayesian Inference techniques being used to plot some confidence intervals around the predictions (Refer: Slide 25). I like to know what Bayesian Optimization technique is being used here. Could you please add the jupyter notebook or python file demo of these plots with uncertainty measures?

rushilanirudh commented 5 years ago

Hi @diliprk thanks for your interest in our work. The confidence intervals shown here do not explicitly use B.I. We were trying to study the confidence in predictions while introducing small perturbations in the latent space of the surrogate. We have developed more sophisticated measures of UQ since, but they are not part of this particular work.