Andrew
summary
Argues for practical Bayesian work by showing what was done: a hierarchical analysis of a sexual‑harassment survey with hundreds of thousands of respondents, modeling doctors’ disciplinary rates versus exam scores for the National Board of Medical Examiners, and retail analyses over millions of transactions. Describes tooling and tradeoffs: using Stan with faster hard‑coded regressions yielding 3–10x speedup, exploring ADVI, expectation propagation, data‑splitting and parallelization, and when to fit simpler models versus accept approximation error.
AI-written summary · about
chapters (12)
- 0:00Bayesian inference basics
- 2:57Science and policy reform
- 6:31Statistics for real-world models
- 8:02Survey analysis case study
- 12:41Why Bayesian methods
- 17:00Software challenges
- 18:04Faster Bayesian computation
- 23:47Statistical computing formalized
- 25:12Big data splitting
- 26:18Closing remarks
- 29:30Engineers and statisticians
- 33:00Q&A
related talks
40mModel Upgrade Schemes- Considerations for Updating Production Models
35mWhat we learned by running a large custom Bayesian...
28mA/B testing and canary releases - why, what and how?
28mHow to manage data-related projects and not fail (too often)?
07mAlexandre Bergeron Interview
09mInterview with
39mDataEngBytes 2023 - BRI-T1-02 - Mikiko Bazeley
35mProduction as an Experiment Lab