Leveraging Conformal Prediction for Calibrated Probabilistic Time Series Forecasts
summary
The talk covers using conformal prediction to produce well-calibrated probabilistic time-series forecasts for energy prices, aiming to help balance grids with volatile renewable supply. It explains conformal prediction (requiring only data exchangeability) and its finite-sample valid coverage, discusses the limitation that standard conformal intervals adapt weakly to local input variability, and shows how to calibrate probabilistic forecasts via methods like conformalized quantile regression.
AI-written summary · about
chapters (16)
- 0:00Conformal prediction intro
- 2:12Renewable energy uncertainty
- 3:40Limits of point forecasts
- 5:36Conformal prediction guide
- 8:53Workflow recap
- 9:33Python packages
- 11:28Prediction intervals drawbacks
- 13:36Calibrating probabilistic forecasts
- 15:28Quantile regression foundation
- 19:37Conformal quantile regression
- 21:49Calibrating each quantile
- 23:48Why CQR solves issues
- 24:31Exchangeability assumption
- 25:35Key learnings
- 26:38Resources and future
- 27:20Q&A
related talks
30mConformal Prediction: uncertainty quantification to humanise models
37mQuantifying Uncertainty in Time Series Forecasting
31mUncertainty quantification: How much can you trust your machine learning model?
36mQuantifying uncertainty in Machine Learning predictions
44mAccuracy Is Not Enough: Building Trustworthy AI with Conformal Prediction
22mConformal Inference with Tidymodels
45mUncertainty estimation at scale with functime
46mPanel about probabilistic forecasting