Brief Tutorial on Probabilistic Databases
summary
Defines a tuple-independent model where each tuple is a random event and query probability equals the probability of the lineage (grounded FO sentence) and reduces evaluation to propositional model counting. Gives a concrete dichotomy: for non-repeating conjunctive queries hierarchical = PTIME and otherwise #P-hard, and extends to unate universal/existential fragments where syntactic rules (independent projection, inclusion–exclusion) are complete. Shows DPLL-style solvers map to decision DNNF and can be provably exponential even when the rules yield polynomial-time evaluation.
AI-written summary · about
chapters (26)
- 0:00Intro
- 0:26About This Talk
- 1:50Background: Relational databases
- 5:20Probabilistic Database
- 8:32Outline
- 9:45Probability of a Formula
- 11:06Algorithms for Model Counting (Gomes, Sabharwal, Selman'2009) Based on full search DPLL
- 13:40Relational Representation (1/2)
- 16:17Weighted FO Model Counting
- 19:27Discussion: Correlations
- 22:33Background: Query Plans
- 23:55An Example
- 26:15Extensional Plans
- 29:50Unsafe Queries
- 33:24Hierarchical Queries
- 34:41The Small Dichotomy Theorem
- 39:14FOun = Unate FO
- 44:27Example: Liftable Query
- 48:42Unliftable Queries (HK)
- 50:04Summary so Far
- 51:01Lifted vs. Grounded Inference
- 52:09Decision-DNNE
- 53:43Notations
- 54:41Easy/Hard Queries
- 55:36Cancellations
- 57:47Möbius Über Alles
related talks
38mThe Intensional-Extensional Problem in Probabilistic Databases
01h 03mAn Overview of Probabilistic Programming
33mCombined Approximations for Probabilistic Query Evaluation: An Intensional Approach
35mApproximating Certain and Possible Answers Beyond Sets
24mHigher order programming with probabilistic effects: A model of stochastic memo...
01h 00mLogic and Probabilistic Circuits 3
01h 05mLogic and Probabilistic Circuits 1
15mScaling Exact Inference for Discrete Probabilistic Programs
