Skip to main content

Uncertainty Quantification for AI Physics: Know When to Trust Your Surrogate Predictions

AI Physics now tells you how much to trust each surrogate model prediction before you use it for engineering decisions. A new Uncertainty section appears inline with every inference result, showing a bounds check that explains exactly which inputs are within or outside the range used for training the model and how closely an ensemble of surrogate models agree on the prediction.

Release Highlights​

What is it?​

A two-layer confidence system built into the inference workflow:

  • Bounds Check - When you run a prediction, the Uncertainty panel shows whether your inputs fall within the range of your training simulations.
  • Ensemble Agreement - Indicates how closely an ensemble of individually trained models agree on the prediction.

Why does it matter?​

  • Avoid bad engineering decisions: Catch unreliable predictions before they propagate into design reviews or manufacturing specs. The inline trust signal makes uncertainty visible at the moment of decision, not as an afterthought.
  • Know when to run full simulations: Use surrogate predictions where confidence is high; validate with full solver runs only where uncertainty flags a risk.
  • Identify training data gaps: The per-parameter coverage chart reveals which regions of the input space are under-represented — so you can target additional simulations where they'll have the most impact.

How is it used?​

When training a model with Auto Tune, select Uncertainty Ensemble to train multiple models that can be used to verify ensemble agreement. When you run inference with this model, you will see an Uncertainty section showing a bounds check and ensemble agreement.

An inference result with an Uncertainty section reading Low Trust, Refine Model, with Bounds Check In Bounds, Ensemble Agreement Low Agreement, and model agreement at 78 percent, beside a 3D pressure field

The Uncertainty section appears inline during inference, showing a bounds check and ensemble agreement

Getting Started​

  1. Navigate to Models in the left sidebar and open a trained surrogate model.

  2. Click the Predict button, select Interactive Prediction, and run inference with your input parameters.

  3. After the prediction returns, scroll to Uncertainty to see the bounds check and ensemble agreement.

For more on building surrogate models and running inference, see the AI Physics documentation.