Personal tools

ProbCog

Probabilistic Cognition for Cognitive Technical Systems

Project Details

Software Suite for Statistical Relational Learning

ProbCog designs and implements a software suite for statistical relational learning that is publicly available for download.

Features:

  • Bayesian Logic Networks (BLNs) [Java implementation]
    • Inference:
      • SAT-based: satisfiability-based importance sampling (SAT-IS) and extensions thereof, MC-SAT
      • Lifted: Backward Simulation
      • standard Bayesian network algorithms (likelihood weighting, backward simulation and variants thereof, EPIS-BN, Gibbs Sampling)
    • Learning: Maximum Likelihood
    • Translation to MLNs
  • Markov Logic Networks [Java and Python implementations]
    • Probabilistic Inference: MC-SAT, Gibbs Sampling
    • MAP inference: MaxWalkSAT, Toulbar2 WCSP algorithms
    • Learning: Maximum Likelihood [Python]
    • Translation of ground Markov random fields to WCSPs
  • Framework for relational data
    • structured collection of relational data
    • scripted generation of relational data
  • Graphical wrappers for inference and learning in BLNs and MLNs with interfaces to external applications (e.g. Alchemy for MLNs)
Prerequisites:
  • supported platforms: Windows (32bit), Linux/i386 and Linux/AMD64 (other platforms may work, but not out-of-the-box; see additional information in the readme file)
  • Java 5 runtime environment (or newer)
  • Python 2.5 (or newer) with Tkinter installed
    Note: Tkinter is part of Python distributions by default on Windows. On Linux, use your package manager (e.g. on Debian, the package is called "python-tk").
Downloads:

Selected Publications

  • Dominik Jain , Paul Maier , Gregor Wylezich, Markov Logic as a Modelling Language for Weighted Constraint Satisfaction Problems, 2009, Eighth International Workshop on Constraint Modelling and Reformulation. In conjunction with CP2009. To appear,
    (BibTeX) (PDF)

Media

Logo
 

Document Actions