BayesDB: Query the Probable Implications of Data
summary
BayesDB claims a usable layer for probabilistic inference over tables and shows an executable system: BQL averages over an implicit space of probabilistic models, runs in notebooks, and ships open source. Demos on a public satellites table infer missing fields, simulate and infer missing values and anomalies, estimate likelihoods (flags unit and decimal errors), and produce confidence scores from an ensemble; plugging in Kepler’s laws and a random forest tightens orbital predictions. Performance is interactive for many queries but some meta‑modeling and simulations remain costly; scales to hundreds of columns.
AI-written summary · about
chapters (14)
- 0:00Introduction
- 0:55The Need for Probabilistic Data
- 3:36BQL: Inferring Implications
- 7:05Bayesian Query Language Overview
- 8:55Satellite Data Demo
- 12:31Data Cleansing and Anomalies
- 14:24Model-Independent Queries
- 16:58Machine-Assisted Modeling
- 21:03Data Correlation vs Prediction
- 23:39Refining and Composing Models
- 26:23Probabilistic Implications Review
- 28:41Future Programming and Tools
- 32:08Q&A: Syntax and Scalability
- 37:22Q&A: Composition and Closing
related talks
40mProbabilistic Programming for Augmented Intelligence
20mGenSQL: A Probabilistic Programming System for Querying Generative Models of Database(…)
42mBayes and Bool Walk into a Bar
47mApproximate Lifted Inference with Probabilistic Databases
39mThe Duct Tape of Heroes: Bayes Rule
38mProbabilistic programming and meta-programming in Clojure
01h 30mBayBE: A Bayesian Back End for Experimental Planning in the Low-To-No-Data Regime
44mBayes' Theorem: The Formula That Changed the World