Extreme Event Likelihoods with Guided Generative Models

Summary

Estimating the likelihood of extreme, high-impact events using brute-force Monte Carlo simulations requires excessive model iterations for expensive physics-based climate models. While guided diffusion models steer generation toward rare events like tropical cyclones, oversampling distorts the true probability distribution. To address this, researchers guide the diffusion-based climate emulator NVIDIA cBottle toward specific events and compute log-odds ratios between guided and unguided distributions. Reweighting the guided samples via these odds ratios enables importance sampling under the original climate distribution. Implemented in NVIDIA Earth2Studio, this technique reduced standard error by 25% compared to simple Monte Carlo sampling when evaluating tropical cyclone states.

Context

Estimating rare, high-impact climate events like tropical cyclones via brute-force Monte Carlo sampling requires large simulation ensembles, creating a bottleneck for risk analysis. Steering diffusion models toward rare events oversamples distribution tails, distorting the original probability.

Approach / What changed

Using NVIDIA cBottle in Earth2Studio, researchers steer a diffusion model toward tropical cyclone states and compute the log-odds ratio between unguided and guided probabilities. This ratio acts as an importance sampling weight to estimate true likelihoods under the base distribution.

Takeaways

• The odds-ratio calculation requires estimating generative flow divergence, which needs second-order derivatives through the diffusion model.
• Combining guided generation with odds-ratio reweighting reduced standard error by 25% compared to simple Monte Carlo sampling in tropical cyclone cases.
• A negative log-odds ratio indicates that the generated sample is substantially more likely under the guided distribution than under the unguided base model.

Related reading

• NVIDIA Developer Blog · Jun 30, 2026

  Optimizing a Neural Reconstruction Pipeline Using NVIDIA Nsight Developer Tools

  NVIDIA Omniverse NuRec reconstructs dynamic 3D environments from camera and lidar sensor data for autonomous vehicle simulation workflows, but high computational demands historically caused long reconstruction times. Engineers used NVIDIA Nsight Systems and Nsight Compute to systematically profile the PyTorch-based training pipeline and optimize its underlying CUDA kernels. Nsight Systems revealed GPU underutilization caused by numerous small kernels and blocking synchronization calls, leading to a fused interpolate kernel that accelerated execution from 4.184 milliseconds to 83.81 microseconds. Profiling with Nsight Compute uncovered resource mismatch in the renderBackward kernel, prompting the team to split it into dedicated camera and lidar implementations with tuned register and shared memory allocations. These adjustments raised kernel occupancy from approximately 15% to between 30% and 50% while decreasing the longest lidar kernel runtime from 31 milliseconds to 18 milliseconds.

  Tanya Lenz

• NVIDIA Developer Blog