Fuzzy Bayesian networks for network inference (Transfer Report)

Abstract: Causal inference is a challenging and important problem in a wide range of scientific fields. Understanding how factors interact is essential in theoretical research, experimental studies and applied research and development.

Causal networks are a particular kind of causal inference problem, where a known number of factors interact with each other in an unknown way. Causal networks range in size from very small (5–10 factors) to very large (104). Although rigorous causal inference is possible in some situations it is much more common to infer a dependency network. This network model of the factors can then be used to guide further direct experimentation which verifies or falsifies suggested causal relationships.

A wide range of techniques have been used to infer causal and network models, including ordinary differential equations, fuzzy and logical networks, neural networks and Bayesian networks. Generally, all of these techniques are intractable on large networks (i.e. networks with more than 100 factors).

In our research we aim to develop techniques which can be used to infer detailed draft models of large causal networks. The techniques will include fuzzy Bayesian networks. Fuzzy Bayesian networks are Bayesian networks that have variables which are simultaneously uncertain and fuzzy. Thus the associated theoretical questions surrounding fuzzy probability are important in our research. This report introduces the problem of causal network inference. It then describes the relevant aspects of a bioinformatic problem domain and previous research. We present a proposed approach to the problem of inferring large causal networks, some bioinformatic data sets that can be used to evaluate and validate this approach, and discuss the goals and contributions. Appendices present our relevant published and current work in the field so far.

  title = 	 {Fuzzy Bayesian Networks for Network Inference},
  author =	 {Christopher Fogelberg},
  howpublished = {Transfer Report, Computing Laboratory},
  address =	 {Wolfson Building, Parks Road, Oxford, OX1-3QD},
  month =	 {October},
  year =	 2008,

Available here.