Bayesian And Causal Network Inference

From OpenCog

Basic Details


The project aims to implement the approach of indefinite probabilities for general artificial intelligence, an interesting method of using ranges of probabilities to model world events. Thus, adding more robustness to sensitivity of parameters. The method then is further implemented in the context of a bayesian network to produce a robust inference mechanism. We will use an extension of the variable elimination algorithm to cope with the particularities of the indefinite probs. Our final goal is to implement a causal network (a mirror-double bayesian network) to answer question based on counterfactuals.



List of Suggestions

Yes: actually, wrapping up some of the indefinite-probabilities/PLN methods in nice library functions could be done at an early stage of this project, and would be a nice side-effect of the work of general value for OpenCog.

Work Log

Week of May 26 (2008, May 26 - Jun 1)

  • May 26 - studied PLN general, in particular Indefinite Probabilities
  • May 27 - studied the opencog IndefiniteTruthValue class implementation
  • May 28 - studying the opencog server and possible integration schemes of Bayesnet in it
  • May 29 - some discussion with mentors about the implementation of PLN
  • May 30 - some opencog hacking for my Mac OS X 10.4
  • Jun 1 - received an email from Joe (new design)

Week of Jun 2 (2008, Jun 2 - Jun 8)

  • Jun 2 - studying the new design ideas
  • Jun 3 - debugging some minor issues of new opencog release on Mac OS X 10.4
  • Jun 4 - added a Bazaar branch in Launchpad
  • Jun 5 - studying C++ templates