Title

From dynamic influence nets to dynamic Bayesian networks: A transformation algorithm

Author Affiliation

Sajjad Haider is Full-time Faculty Member at Institute of Business Administration (IBA), Karachi

Faculty / School

Faculty of Computer Sciences (FCS)

Department

Department of Computer Science

Was this content written or created while at IBA?

Yes

Document Type

Article

Source Publication

International Journal of Intelligent Systems

ISSN

0884-8173

Abstract

This paper presents an algorithm to transform a dynamic influence net (DIN) into a dynamic Bayesian network (DBN). The transformation aims to bring the best of both probabilistic reasoning paradigms. The advantages of DINs lie in their ability to represent causal and time-varying information in a compact and easy-to-understand manner. They facilitate a system modeler in connecting a set of desired effects and a set of actionable events through a series of dynamically changing cause and effect relationships. The resultant probabilistic model is then used to analyze different courses of action in terms of their effectiveness to achieve the desired effect(s). The major drawback of DINs is their inability to incorporate evidence that arrive during the execution of a course of action (COA). Several belief-updating algorithms, on the other hand, have been developed for DBNs that enable a system modeler to insert evidence in dynamic probabilistic models. Dynamic Bayesian networks, however, suffer from the intractability of knowledge acquisition. The presented transformation algorithm combines the advantages of both DINs and DBNs. It enables a system analyst to capture a complex situation using a DIN and pick the best (or close-to-best) COA that maximizes the likelihood of achieving the desired effect. During the execution, if evidence becomes available, the DIN is converted into an equivalent DBN and beliefs of other nodes in the network are updated. If required, the selected COA can be revised on the basis of the recently received evidence. The presented methodology is applicable in domains requiring strategic level decision making in highly complex situations, such as war games, real-time strategy video games, and business simulation games.

Indexing Information

HJRS - W Category, Scopus, Web of Science - Science Citation Index Expanded (SCI)

Publication Status

Published

Share

COinS