Overview

The Bayes-Knowledge research team uses Bayesian Networks to assess risk and aid decision-making in a wide range of applications.

New to this?

You have a decision to make, but where do you start?

Most real-world decisions must strike a balance between limited data and professional conjectures, in a workplace that won't stand still.

Bayes-Knowledge methods bring together data, expert opinion, risk and uncertainty into a formal statistical framework, allowing decision-makers to see their choices clearly.

A little more depth

In the Bayes-Knowledge research team we develop novel strategies for Bayesian networks, using new algorithms to extend and refine the use of BNs in decision-making.

Bayesian Networks: greater than the sum of its parts

Bayesian networks are built on a structure of causes and effects, into which information of many types may be integrated. They may be used in the same way as classical methods, where data is the primary source of information- however, BNs can also handle those scenarios where data are sparse, or unreliable, by allowing other sources of information to be tapped. This can be crucial when a situation is constantly changing, such as when data collected in the past are not useful for predicting what is likely in the future.

By using BNs, data need not be divorced from their context, which is useful for interpreting findings: a likely causal structure, expert opinions, data past and present, and expressions of uncertainty are all brought together- creating a natural platform for pragmatic decision-making. Furthermore, the process of building a BN requires all assumptions to be stated clearly, which is key for dealing with uncertainty fully, and choosing appropriate sensitivity analyses.

Bayes-Knowledge: the nuts-and-bolts, or what's new?

Using information smartly is the fastest path to the best decision: "dynamic discretization" is a tool for BNs created by the Bayes-Knowledge project, which allows the necessary discretization of continuous variables to be optimised through an iterative process. This allows us to model every variable in the BN with minimum inefficiency and maximum accuracy, and probability distributions for continuous variables need not be constrained to a convenient shape.

A valuable application of dynamic discretization in the Bayes-Knowledge project has been to improve methods for statistical inference in BNs, which we have dubbed "parameter learning".

Bayes-Knowledge in the real world

In the Bayes-Knowledge project we believe in applying our theories to real-world problems. Our methods may be used in any decision-making situation, and we have published applications for a diverse range of areas, as can be seen from our list of publications.

The methods and improvements developed in the Bayes-Knowledge project are implemented in AgenaRisk, which is available for PC, Mac and Linux.

The Bayes-Knowledge project is funded through the European Research Council, and is based at the Electronic Engineering and Computer Science department at Queen Mary University London.

Summary of the Project

 

We use cookies to improve your experience and to help us improve our site. Continuing to use this service means you agree with our Cookies policy.