Guaranteed prediction sets for functional surrogate models
This addresses the need for reliable uncertainty quantification in scientific machine learning, particularly for PDE emulators, but is incremental as it builds on existing conformal prediction and set-propagation techniques.
The authors tackled the problem of building reliable PDE emulators by proposing a method to obtain statistically guaranteed prediction sets for functional surrogate models, resulting in prediction sets with conformal prediction coverage guarantees.
We propose a method for obtaining statistically guaranteed prediction sets for functional machine learning methods: surrogate models which map between function spaces, motivated by the need to build reliable PDE emulators. The method constructs nested prediction sets on a low-dimensional representation (an SVD) of the surrogate model's error, and then maps these sets to the prediction space using set-propagation techniques. This results in prediction sets for functional surrogate models with conformal prediction coverage guarantees. We use zonotopes as basis of the set construction, which allow an exact linear propagation and are closed under Cartesian products, making them well-suited to this high-dimensional problem. The method is model agnostic and can thus be applied to complex Sci-ML models, including Neural Operators, but also in simpler settings. We also introduce a technique to capture the truncation error of the SVD, preserving the guarantees of the method.