New Release - Hugin 6.4

On March 2nd, 2004 - we have released new versions of the Hugin Graphical User Interface (v6.4) and Hugin Decision Engine (v6.2).

The main new features of this release are:

• The Hugin Graphical User Interface now includes a tool for analysing dependence and independence relations between nodes given evidence (d-separation)
• Automatic update of Hugin Graphical User Interface (receive and install updates automatically)
• Compressed and password protected hkb-files
• Various improvements of Hugin Graphical User Interface including a substantial improvement of the speed of the Hugin Graphical User Interface
• The Hugin Decision Engine now has support for EM learning in object-oriented Bayesian networks

Hugin Graphical User Interface v6.4:
A tool for performing dependence and independence analysis between nodes of a probabilistic graphical model (i.e. Bayesian network or influence diagram) has been included in the Hugin Graphical User Interface. This tool enables the user to perform d-separation analysis on sets of variables given selected sets of evidence variables. This tool is particularly useful in the knowledge acquisition phase for validating the dependence and independence properties of the model.

The Hugin Graphical User Interface now supports live updating. This feature - when enabled - will make the Hugin Graphical User Interface automatically check for available updates from our web site. Updates of the Hugin Graphical User Interface are available from http://www.hugin.com/Products_Services/Products/Update/.

The Hugin Graphical User Interface now uses compressed hkb-files (Hugin Knowledge Base files). Furthermore, password protected hkb-files is now an option. Password protection of hkb-files is useful when models are distributed as part of an end-user application.

The Hugin Graphical User Interface has been improved with various new features. This includes:

• Considerable efforts have been put into improving the speed of the Hugin Graphical User Interface.
• "Live" node resize. This feature enables the user to resize nodes by dragging.
• It is possible to view the fading, experience, and expression tables for a node at the same time as she is viewing the conditional probability table of the node.
• Parameter learning and adaptation for discrete chance nodes in influence diagrams and mixed Bayesian networks is now supported by the Hugin Graphical User Interface.
• Simulation of node instantiations (i.e., generation of cases) is now supported for mixed Bayesian networks.
• The Hugin Graphical User Interface now has improved support for locating nodes in large models. This feature enables the user to search for a node in a large network by its name.
• The user can select to show node names and/or labels in the nodes.
• The user can set the precision used to display the entries of a table.
• It is possible to print a junction forest to a file or a printer.
• Monitor resizing. This feature enables the user to resize node monitors for better display of results of propagation.
• It is possible to set the number of samples per interval used by the Table Generator when generating the table for a discrete node given parent interval nodes.
• The user can select to view either node state labels or values in conditional probability tables.
• The Hugin Graphical User Interface now has better support for navigating instance nodes and entering evidence on nodes in instance nodes of a compiled network. The user has the option to transfer evidence from a compiled subnet to a corresponding instance in a different compiled network.

Hugin Decision Engine v6.2:
The Hugin Decision Engine now has support for parameter learning in object-oriented Bayesian networks using the EM (Expectation Maximization) algorithm. This feature enables the user to exploit the composition of an object-oriented Bayesian network when estimating conditional probability tables from data.

The Hugin Decision Engine now uses compressed hkb-files (Hugin Knowledge Base files). Furthermore, password protected hkb-files is now an option. Password protection of hkb-files is useful when models are distributed as part of an end-user application.

The Hugin Decision Engine supports parsing a set of nodes from a file. This is, for instance, useful for loading triangulations from a file. The Hugin Decision Engine also supports parsing a database of cases stored in an ASCII text file. This is useful when learning Bayesian networks from data.