Kristian Kersting, joint work with Luc De Raedt

1
Partially supported by EU IST programme under
contract number IST-2001-33053 APrIL -
Application of Probabilistic Inductive Logic
Programming
Representional Power of Probabilistic-Logical
Models
From Upgrading to Downgrading
SRL 2003, Acapulco, Mexico, August 2003

• Kristian Kersting, joint work with Luc De Raedt
• University of Freiburg

2
Overview

• Probabilistic Logic Learning
• Upgrading
• Downgrading
• Applications

3
What is Probabilistic Logic Learning?
Survey in SIGKDD Explorations special issue on
MRDM, in press
Representations and reasoning mechanisms grounded
in probability theory, e.g. HMMs, Bayesian
networks, stochastic grammars,... Successfully
used in a wide range of applications such as
Robust models
Logic programming Elegant representation of
complex situations involving a variety of objects
as well as relations among these objects
bloodtype(X,a) lt- mother(M,X),bloodtype(M,a), fat
her(F,X),bloodtype(F,a).
Logic
Probabilistic

• Computational biology
• Speech recognition

Learning
Often it is easier to obtain data and to learn a
model than using traditional knowledge
engineering techniques

• Parameter estimation
• Learning the underlying logic program
• Fully vs. not observable random variables

4
Upgrading (1,2,3,4)
5
Upgrading (...)

• Cyclic, acyclic
• Finite, discrete, continuous random variables
• ...

6
Overall Scientific Objective
One of the key open questions of artificial
intelligence concerns "probabilisitic logic
learning",
i.e. the integration of probabilistic reasoning
with first order logic representations and
machine learning.
Logic
Probabilistic
Learning
7
Downgrading

• Choose generally applicable PLM
• Downgrade to strike the right balance between
  expressivity and learnability.

Example Logical Hidden Markov Models (LOHMM)
Kersting et al. 02
Stochastic Logic Programs
iterative clauses
8
Advantages

• Guarantee of unique probability measure
• No particular PLM is favoured
• Impact of language concepts ?
• General understanding of PLM / PLL ?
• Shift from representations to applications

9
Applications ?

• Other forms of observabilty ?
• contact(?,kristian) true contact(?,kristian)
  ?
• Long-term correlations due to memory/functors
• Explicit modelling of logical constraints
• pc(X) mother(Y,X) , pc(Y), mc(Y).
• Relational SVMs due to Fisher Kernels
• Tabling
• Magic sets
• Background knowledge, language search bias

10
Thanks !