Examples

End-to-end examples on real datasets, as notebooks: read them here, or run them on Google Colab with one click. Each one explains a single uncertainty technique, applied either during training or after training (post-processing).

Tasks

• Forecasting: Forecast the temperature 24 hours ahead: at a weather station from its last 48 hours of measurements, or everywhere on Earth with a pre-trained global weather model. (4 examples)
• Regression: Predict house values in California districts from census data. (2 examples)
• Classification: Classify clothing images, and recognize inputs unlike the training data (handwritten digits). (4 examples)

All examples

• Bayesian weather forecasting: what the model knows and what it doesn't — Forecasting, training
• Probabilistic weather forecasting with a Gaussian head — Forecasting, training
• Conformal prediction intervals for a weather forecaster — Forecasting, post-processing
• Probabilistic weather forecasts from a pre-trained FourCastNet — Forecasting, post-processing
• House prices with a mixture density network — Regression, training
• Adaptive conformal intervals for house prices — Regression, post-processing
• A Bayesian CNN: asking a network several times — Classification, training
• Detecting unfamiliar inputs with SNGP — Classification, training
• Conformal prediction sets: "it's one of these" — Classification, post-processing
• Fixing an overconfident classifier with temperature scaling — Classification, post-processing