SIGMA: Uncertainty-aware validation of generative models
What it is
SIGMA is an end-to-end framework that combines density estimation using Gaussian mixture models with principled uncertainty quantification. It constructs a new goodness-of-fit test that profiles model fitting uncertainties as nuisance parameters, allowing for the validation of generative models on larger test sets without falsely rejecting them due to estimation error.
The problem it solves
Standard goodness-of-fit tests treat generative models as exact, leading to false rejections when test sets are large enough to resolve the statistical error inherent in fitting to a finite training set.
What you could use it for
- Validating background estimation models in particle physics
- Assessing the reliability of generative models in high-energy physics benchmarks
- Performing statistical tests on density estimates where training data is finite
Why it matters
Enables the robust use of generative models in physical sciences by distinguishing between genuine mismodeling and statistical estimation error, thereby allowing models to be trusted on significantly larger test sets.
Read the source ↗Back to the library →
Summary written by an AI model from the source; check the source before you rely on it.