GenSQL: A Probabilistic Programming System for Querying Generative Models of Database(…)
summary
GenSQL is a SQL extension that runs Bayesian queries over generative table models, provides a unified model API (supports PPLs, GLMs, CTGAN, and a crosscat-like structural learning backend), and proves soundness with a typed denotational semantics. It powers clinical-trial anomaly detection and wet-lab conditional synthetic data, delivers 1.7–6.8× speedup versus a specialist DB and shows only 1.1–1.7× overhead versus hand-written code, and supports structural program synthesis to auto-learn usable generative models.
AI-written summary · about
chapters (9)
- 0:00GenSQL introduction
- 1:48System overview
- 3:36Language constructs
- 7:53Query processor workflow
- 10:43Generative model sources
- 12:09Case studies: biology & medicine
- 14:14Performance benchmarks
- 16:20Formal semantics
- 18:48Future directions
related talks
42mBayesDB: Query the Probable Implications of Data
17mThe keys to success with Generative AI and Vector Databases
47mApproximate Lifted Inference with Probabilistic Databases
37mRound Table Discussion
18mFast Inference for Probabilistic Graphical Models
37mBreakdown New Data Silos in Generative AI
13mggsql: A grammar of graphics for SQL
35mBreakout Session