Concepts
Concepts: Uncertainty in Quantum Machine Learning
Quantum models introduce several sources of uncertainty:
- Shot noise (statistical): finite sampling of measurement outcomes.
- Hardware noise (aleatoric): decoherence, gate errors, readout noise.
- Model uncertainty (epistemic): limited training data, model misspecification.
QuantumUQ focuses on model-agnostic techniques that sit on top of existing QML models:
- ShotBootstrap: resample shots / repeated forward passes.
- DeepEnsemble: independent predictors trained from different initializations or data splits.
- NoiseProfile: sweep shots and quantify stability (entropy and probability variance).
All methods work on a small Predictor protocol (predict, predict_proba, and task).