Quantifying uncertainty in Machine Learning predictions
summary
Claim: conformal prediction gives guaranteed error bounds (1 − ε) with almost no assumptions and works with any model, few assumptions. Method: split training/calibration, pick a nonconformity score, compute p‑values and output label sets or prediction intervals. Evidence: applied to an insurance classifier and a regression price model, yields calibrated probabilities and a reported 67% reduction in false positives. Slides are linked in the uploader description.
AI-written summary · about
chapters (10)
- 0:00Uncertainty in ML predictions
- 1:09Conformal prediction example
- 5:08Conformal prediction assumptions
- 8:01Conformal prediction intuition
- 10:31Formal conformal predictor
- 14:48Implementation guide
- 17:19LV Insurance classification
- 23:33LV Insurance regression
- 27:33Key takeaways
- 30:26Q&A
related talks
31mUncertainty quantification: How much can you trust your machine learning model?
35mUncertainty Unleashed
30mConformal Prediction: uncertainty quantification to humanise models
44mAccuracy Is Not Enough: Building Trustworthy AI with Conformal Prediction
37mQuantifying Uncertainty in Time Series Forecasting
44mHandling predictive uncertainty in Machine Learning
13mDistribution-Free, Risk-Controlling Prediction Sets
32mLeveraging Conformal Prediction for Calibrated Probabilistic Time Series Forecasts