| Legacy
Legacy

StaRAI or StaRDB?

A tutorial on Statistical Relational AI (StaRAI) at BTW 2019
18. Fachtagung für "Datenbanksysteme für Business, Technologie und Web, 4-8 March 2019, Rostock, Germany

Abstract

In recent years, a need for compact representations of large relational databases became apparent, e.g., in natural language understanding or decision making. Using inductive logic programming (ILP), one can build a model of a database, allowing for a crisp reproduction of data. Another idea is to build a so called factor graph model of data and introduce a probability distribution to reproduce data approximately. Such a model defines an intensional representation of a probabilistic database. A factor graph model uses parameterised variables similarly to the variables in ILP to compactly represent relations and objects. Grounding such a model incurs an exponential blowup and makes inference infeasible. Instead of grounding out a model, one can answer queries on the model directly and in a scalable way.

This tutorial aims at connecting databases and StarAI, demonstrating how database systems can benefit from methods developed within StarAI, e.g., for implementing effi- cient systems combining databases and StarAI. Thus, the goal of this tutorial is two-fold:

  1. Present an overview of methods within StarAI.
  2. Provide a forum to members of both communities for exchanging ideas.

An accompanying proceedings article can be found here.

Presenter

  • Tanya Braun (University of Lübeck at the time)

Outline

  1. Introduction (pdf)

  2. Probabilistic relational modeling (pdf)

    • Semantics

    • Inference problems and applications

    • Algorithms and systems

    • Scalability

  3. Scalability by lifting

    • Exact lifted inference (pdf)

    • Approximate lifted inference (pdf)

  4. Summary (pdf)