TechPrufer
Sign inGet started

Blogs›NVIDIA›Guided Generative Models for Rare Event Likelihood Estimation

Guided Generative Models for Rare Event Likelihood Estimation

Guided Generative Models for Rare Event Likelihood Estimation

1
posts
2026

This feature thread tracks the development and application of guided generative models, particularly diffusion models, for efficiently estimating the likelihood of rare, high-impact events. Initial work focused on climate science, using models like NVIDIA cBottle and tools like Earth2Studio, to steer generative processes towards extreme weather phenomena (e.g., tropical cyclones) and then employing odds-ratio diagnostics to correct for sampling bias. Subsequent developments have demonstrated the application of this methodology to tropical cyclone risk estimation, achieving significant standard error reduction compared to traditional Monte Carlo methods by combining guided generation with odds-ratio diagnostics. Future work is focused on optimizing computational efficiency, improving density estimation stability, and extending guided sampling techniques to a wider range of extreme phenomena and attribution studies.

2026

Extreme Event Likelihoods with Guided Generative Models | NVIDIA Technical Blog

7/13/2026

This post details the practical application of guided diffusion models for estimating the likelihood of rare extreme events, specifically tropical cyclones, using NVIDIA cBottle and Earth2Studio. It explains the concept of using odds ratios to correct for oversampling introduced by guidance, enabling importance sampling of rare events while maintaining accurate probability estimation under the original climate distribution. The post provides a minimal working example with Python code demonstrating how to run guided sampling and calculate the log-odds ratio, highlighting a 25% standard error reduction in tropical cyclone likelihood estimation compared to simple Monte Carlo sampling.