Math & Optimization · advanced · concept 71 of 311 · Updated Nov 4, 2026
Bayesian ML & Uncertainty
Instead of one best model, keep a distribution over models and let predictions carry error bars. Bayesian methods quantify what the model does not know, which matters when a wrong-but-confident answer costs money or lives. The ideas power probabilistic programming and calibrated forecasting.
Top AI contributors on this
- Michael I. JordanInria Paris / UC Berkeley (emeritus)
- John HopfieldPrinceton University (emeritus)
- Cassie KozyrkovIndependent (formerly Google)
- Chelsea FinnStanford University / Physical Intelligence
- Daphne Kollerinsitro
- Sébastien BubeckOpenAI
Key terms
Posterior distributionUncertainty quantificationMCMCProbabilistic programmingCalibration
Learn these first
Where you meet it in the real world
Drug trials, demand forecasting with error bars, active learning, autonomous-system safety cases
Videos
▶ Bayes theorem, the geometry of changing beliefs ↗
3Blue1Brown · YouTube
Guides and articles
Probabilistic Machine Learning ↗
Kevin Murphy
18. Gaussian Processes: Dive into Deep Learning 1.0.3 documentation ↗
Dive into Deep Learning
Courses, papers, and more
This unlocks