Research at Bayesian Intelligence
Bayesian network technology is rapidly growing, both in the underlying
technology itself and in applications throughout the community.
It is necessary to maintain connection with continuing research
in the technology to deliver the best possible services in their
use. People at Bayesian Intelligence are actively engaged in
relevant research, and espcially in:
- Knowledge Engineering means eliciting information from
human experts to build representations useful for modelling systems or
problems of interest. In addition to simply consulting with
experts, knowledge engineers have access to a wide variety
of specialised techniques, such as
- computer programs to assist with eliciting probabilities,
- constraint optimisation programs for finding the closest probabilities to those
elicited which are consistent,
- programs for performing sensitivity analysis on Bayesian networks
to help validate or correct elicited probabilities and causal structure,
- measures and programs for evaluating Bayesian networks using
empirical data,
- methodologies for Knowledge Engineering Bayesian Networks (KEBN).
We conduct research into all of these issues.
- Data Mining is the use of computational techniques for
discovering patterns in large volumes of data, which used to be known
as machine learning. There are a large number of types of representation
of such patterns, including:
- probability distributions
- mixtures of probability distributions
- sequences of distributions (e.g., for time series analysis)
- classification trees and graphs
- Bayesian networks
- naive Bayesian networks (simplified, for prediction problems)
- dynamic Bayesian networks (complex BNs for time series or planning)
We have a research history with all of these forms of representation
and using a variety of data mining techniques, including: causal
discovery search algorithms (CaMML, PC, GES, K2); genetic algorithms
for search; MML, orthodox, and other methods for measuring the value
of a representation. We also have special expertise in the evaluation
of data mining techniques.