< Back to previous page

Project

Probabilistic Logic and Functional Programming Synergies

Probabilistic graphical models (PGMs) have been popular in artificial intelligence and machine learning since the late 80s. Today, they enable numerous applications in domains ranging from robotics to natural language processing and bio-informatics. However, PGMs essentially define a joint probability distribution of a fixed and finite set of variables, which severely limits their expressivity.

In the pursuit of a universal representation to deal with a variable number of objects and relationships, two approaches emerged:

  1. a possible world semantics for probabilistic logics, and
  2. a semantics based on probabilistic execution traces for probabilistic programming languages (PP).

The two branches are now studied almost in isolation and have quite different focuses. PP considers functional and imperative programs, continuous random variables, Monte-Carlo inference, and Bayesian machine learning. In contrast, statistical relational artificial intelligence focuses on logical and database representations, discrete distributions, knowledge compilation and lifted inference, and structure learning.

This project seizes the opportunity for cross-fertilisation posed by the limited mutual awareness of the two communities: with our rich experience in both probabilistic logic programming and in functional programming, we exploit the synergy between the two branches in terms of semantics, analysis, meta-programming and inference for mixed discrete/continuous distributions.

Date:1 Jan 2017 →  31 Dec 2020
Keywords:Probabilistic Logic Programming, Functional Programming
Disciplines:Applied mathematics in specific fields