Lecture 15: Probabilistic Programming

Lecture 15: Probabilistic Programming¶

AIMA Chapter 15 — 1 hour¶

Learning Objectives¶

• Understand relational probability models

• Model open-universe probability

• Use programs as probability models

• Apply probabilistic programming languages

Relational Probability Models¶

• Objects: Variable number

• Relations: Between objects

• Parameter tying: Share parameters across objects

Open-Universe Models¶

• Unknown objects: Citation matching, entity resolution

• Uncertain identity: Is X same as Y?

• Inference: MCMC, particle filtering

Programs as Probability Models¶

• Generative: Program samples world

• Inference: Condition on observations

• Probabilistic programming: Church, Stan, Pyro

Summary¶

• Relational: Objects, relations, parameter tying

• Open universe: Unknown objects, identity uncertainty

• Generative programs: Sample and condition

• PPLs: Church, Stan, Pyro—inference automation

References¶

• Russell & Norvig, AIMA 4e, Ch. 15

• Chapter PDF: chapters/chapter-15.pdf

Questions?¶

Next lecture: Making Simple Decisions (Chapter 16)