Boosting AI Reliability: Uncertainty Quantification with MAPIE
summary
Model-agnostic MAPIE produces conformal prediction sets with formal coverage guarantees (eg 95% coverage claimed) by splitting into train/calibration/test and computing nonconformity quantiles; it implements split, CQR and Mondrian group-wise variants, supports rolling updates/partial_fit for time series change‑point detection and reports rolling coverage, and plugs into scikit-learn TensorFlow PyTorch models. Library is on GitHub, has notebooks, and a 1.0.0 milestone; demos show intervals widen on drift so the approach visibly detects failures.
AI-written summary · about
chapters (9)
- 0:00AI reliability intro
- 3:56Uncertainty for AI risk
- 7:30Conformal prediction
- 14:46MAPIE library
- 19:44MAPIE roadmap
- 20:46Time series demo
- 26:12Computer vision demo
- 28:40Reliable AI future
- 30:24Q&A
related talks
30mBoosting AI Reliability- Uncertainty Quantification with MAPIE
01h 24mBoosting AI Reliability: Uncertainty Quantification with MAPIE
30mConformal Prediction: uncertainty quantification to humanise models
44mAccuracy Is Not Enough: Building Trustworthy AI with Conformal Prediction
35mUncertainty Unleashed
34mPrediction intervals
36mQuantifying uncertainty in Machine Learning predictions
19mUnderstanding, Generating, and Evaluating Prediction Intervals