The Hugin Graphical User Interface

What It Is

The Hugin Graphical User Interface is an interactive tool enabling you to use the facilities of the Hugin Decision Engine. It can help you construct models that can be used in other applications.

The Hugin Graphical User Interface is also ideal for educational purposes. When introducing the concept of Bayesian networks to a group of students, they will be very motivated if they can model and test Bayesian networks using an easy-to-use tool.

The Hugin Development Environment

The Hugin Graphical User Interface is a component of the Hugin Development Environment. The two other main components are the Hugin Decision Engine and the Hugin Application Programming Interfaces.

Basic Concepts

Before you can use the Hugin Graphical User Interface, you should at least understand the concept of Bayesian networks which is described in the Tutorials section. This section also contains a step-by-step description of how to construct a Bayesian network using the Hugin Graphical User Interface.

The extension of Bayesian networks with decision and utility nodes, known as influence diagrams, allows you to model decision scenarios explicitly. If you are not familiar with (limited-memory) influence diagrams (LIMIDs), you can also learn about these in the Tutorials section. There is also a step-by-step description of how to construct a LIMID using the Hugin Graphical User Interface.

Also, you can learn about the concept of object-oriented networks, which provides a very powerful mechanism for constructing models with repetitive patterns and for constructing models in a hierarchical fashion (top-down, bottom-up, or a mix of the two), making large models much more readable. Again, there is a step-by-step description of how to construct an object-oriented network using the Hugin Graphical User Interface.

Learning More About Hugin

Familiarize yourself with node tables, the table generator, how to perform structure learning, learning of CPTs, how to adapt the parameters of a network with experience, case generator, net language.

When you have familiarized yourself with the most basic functionality of the Hugin Graphical User Interface, you can start using the more advanced features:

The node table tutorial explains the functionalities of node tables.
The table generator tutorial shows how to specify simple expressions for large tables and then let the built-in table generator do all the hard work of filling in the numbers of the table.
The structure learning tutorial describes how Bayesian networks can be constructed automatically from data.
The EM learning tutorial describes how the probabilities (parameters) of Bayesian networks can be learned automatically from data.
The adaptation tutorial explains how the probabilities specified for Bayesian networks can be automatically updated from experience (i.e., evidence) such that, for example, the networks adapt to changing conditions in its environment.
The case generator tutorial explains how to generate simulated cases from a Bayesian network.

In the Hugin APIs section, you can read about the opportunity to include Hugin Bayesian networks and influence diagrams in C, C++, Java, or Visual Basic programs.

You can get a detailed view of the components of the Hugin Development Environment by reading the Components Description.

The Origin of Hugin

During an EU sponsored research project (under the ESPRIT program) on diagnosing neuromuscular diseases, the Bayesian network MUNIN was constructed. A research group at Aalborg University worked on developing correct and efficient computation methods for the diagnosis problem. Some results had at that time been obtained by American researchers, but a very obstinate problem still remained, which prevented Bayesian networks from being used in the construction of expert systems. The problem was known as the rumour problem: you may hear the same story through several different channels; but still the story may originate from the same source. Without knowing whether or not your channels are independent, you cannot combine the statements correctly.

In Bayesian networks, the rumour problem appears when a cause can influence the same event through different paths in the network.

The problem was solved and general methods were made available to be used in any domain which can be modeled by a Bayesian network.

The methods were programmed into a general development and runtime system, which was easy to use for anyone wishing to construct an expert system based on Bayesian networks. The system was called Hugin. Over the years the system has been extended in various ways (e.g. (limited-memory) influence diagrams (LIMIDs), continuous variables, structure learning, adaptation, object-oriented specification of Bayesian networks and LIMIDs, etc).

Hugin Expert A/S

Back