Current Projects

• Bayesian Poker. This now has its own page!

• Bureau of Meteorology: Investigating the use of Bayesian Networks to support weather forecasting. A prototype project in 2000 used the additional data gathering associated with the Sydney Olympics to improve seabreeze prediction. The study used both expert elicitation and CaMML (Causal Discovery via MML) to find Bnets that improved upon existing BOM prediction systems. Current work is applying Bnets to weight and integrate predictions from different numerical programs modeling distinct weather processes.
  

  Participants: Ann Nicholson, Kevin Korb, Tali Boneh (Ph.D. student), Chris Ryan (BOM), John Bally (BOM), James Kelly (BOM).
  

  Publications:

  • R J Kennett, K B Korb and A E Nicholson (2001) Seabreeze prediction using Bayesian networks. In D Cheung, G J Williams and Q Li (eds), Proceedings of the 5th Pacific-Asia Conference on Advances in Knowledge Discovery and Data Mining - PAKDD 2001. (pp 148-153). Springer-Verlag.
  • R Kennett, K B Korb and A E Nicholson (2001) Seabreeze prediction using Bayesian networks: A case study, School of Computer Science and Software Engineering, Monash University, Melbourne, 12pp. Technical report 2001/86.

• SARBayes: Bayesian Models for Search & Rescue We are developing technology to assist with the prediction of the behavior of lost persons. There are two main aspects of work. First, the creation of a large and ongoing database of Australian cases, to extend and perhaps contrast with U.S. databases. Second, the development of Bayesian nets which improve predictions over existing models, such as Syrotuck's, both by getting the distance right more often and by better tracking their own uncertainty. WE are also designing and implementing a general-purpose, open-source library for optimum resource allocation for SAR.
  

  Participants: Charles Twardy, Rik Head (Emergency Systems Technology), Victorian Police Search & Rescue, Kevin Korb.
  

  Publications:

  • Adam Golding (2001) SARBayes. Honours Thesis, School of Computer Science, Monash University.

• Knowledge Engineering a Bayesian Network for an Ecological Application (2003 Honours project)

  This project involves the development of a methodology and software tools to support the task of knowledge engineering a Bayesian network. The KEBN methodology and tool will be developed and evaluated in the context of a particular application area, namely providing intelligent decision support for the assessment of ecological risk associated with irrigation systems in the Goulburn-Broken catchment. This Honours project is being undertaken in conjunction with the Dr Carmel Pollino and other members of the Water Studies Center, Monash University, who are currently working on this ecological risk assessment problem as part of the NPIRD Irrigation ERA Project. The first stage of this project has been the analysis and 'reverse engineering' of the BN developed by Dr. Pollino in 2002. A number of possible improvements have been identified: combining multiple networks into a single network that includes site and type "spatial" information; explicitly including temporal information for predicting over different time frames; and identification of suitable evaluation measures and applying them on the existing and subsequent models.

  Participants: Ann Nicholson, Kevin Korb and Owen Woodberry (Honours student), Carmel Pollino and Mike Grace (Monash University Water Studies Centre).
  

• Using Bayesian networks to predict recreational water quality in Sydney Harbour. (2003 Honours project)

  It is well established that degradation of recreational water quality through faecal contamination has a deleterious effect on the health of swimmers. Since most pathogens are not easily detected in water, indicator bacteria are widely used in the monitoring of recreational water bodies. The principal source of faecal contamination is stormwater runoff and sewage overflows. The NSW Environment Protection Authority has been monitoring beaches in Sydney Harbour for several years by taking water samples and recording bacteria counts from each site. The data available is noisy, limited however replication of samples over time does show interesting patterns. Some preliminary analyses carried out by Dr Grant Hose, of UTS, Sydney, using cluster analysis and non-metric MDS, show that sites distant from the harbour mouth group together, and have higher frequencies of exceeding water quality guidelines, while those closer to the harbour mouth have lower frequencies of exceeding water quality guidelines. Linear regression relationships using lagged rainfall as predictors also generally gave better predictions of faecal and enterococci counts (log transformed) at sites further away from the harbour mouth. The NSW EPA is interested in developing models that would predict levels of bacterial contamination and thus suitability for recreational activities. Current reporting of beaches is only retrospective, so there is no linkage between current environmental conditions and recreational water quality.

  As causal relationships are in principle well understood but difficult to quantify, Bayesian Belief Networks could provide a useful approach to the currently available data. Comparisons with simple univariate lagged regression models would be able to provide measures of the relative effciency of the network approach compared with a more traditional statistical approach. It should be noted that the bacterial modelling of harbour / estuarine systems has received little attention, as most previous work has focussed on ocean beaches. However, the complexity of the factors influencing harbour pollution situation make it a more interesting problem and one which might well highlight the advantages of a Bayesian Network approach.

  In addition, should a Bayesian Belief Network approach prove to be promising, other data sets could perhaps be incorporated. Flushing of the harbour would be related to tidal cycles, easily available, while information on spatial flushing patterns could be obtained from circulation models developed for Sydney Harbour (Das et al). Some sewerage upgrades have been undertaken over the last few years, so exploration of the data might reveal whether such upgrades have been effective in reducing pollution levels. Details of stormwater outfalls and dates and locations of sewer upgrades by catchment should be available from Sydney Water. It is anticipated that this current honours project could form the basis of subsequent postgradate research addressing important and challenging environmental problems.

  Participants: Charles Twardy, Ann Nicholson and Shannon Watson (Honours student), Geoff Gordon, Tim Pritchard (Ecotoxicology and Water Science, NSW EPA.) Grant Hose, (Institute for Water and Environmental Resource Management, UTS).
  

• User Modeling with Bayesian Networks We are using Bayesian networks to model humans in a number of different contexts:
  • The NAG (Nice Argument Generator) architecture for understanding and generating arguments uses Bayesian networks both for normative domain modeling and for modeling users.
  • Keyhole Plan Recognition applies Bayesian nets to identifying plans adopted by users in an adventure game setting.
  • Pre-sending information on the WWW. Bayesian nets are applied to better predict which web pages the user will attempt to retrieve next, which can be used to improve efficiency in handling web traffic.
  • Intelligent tutoring systems. The use of Bayesian networks to model student misunderstanding of decimals.
  
  

  Participants: Ann Nicholson, David Albrecht, Ingrid Zukerman, Kevin Korb, Tali Boneh.
  

  Publications:

  • K Stacey, L Sonenberg, A Nicholson, T Boneh and V Steinle (2003). A teaching model exploiting cognitive conflict driven by a Bayesian network. To appear in Proceedings of the Nineth International Conference on User Modeling .
  • A E Nicholson, T Boneh, T Wilkin, K Stacey, L Xonenberg and V Steinle (2001). A Case Study in Knowledge Discovery and Elicitation in an Intelligent Tutoring Application. In J Breese (ed), Proceedings of the Seventeenth Conference on Uncertainty in Artificial Intelligence, pp 386-394. Morgan Kaufmann.
  • Zukerman, R McConachy, K B Korb and D Pickett (1999). Exploratory interaction with a bayesian argumentation system. In T Dean (ed), Proceedings of the Sixteenth International Joint Conference on Artificial Intelligence (IJCAI-99), pp 1294-1299. Morgan Kaufmann.
  • D W Albrecht, I Zukerman and A E Nicholson (1999). Pre-sending documents on the WWW: a comparative study. In T Dean (ed), Proceedings of the Sixteenth International Joint Conference on Artificial Intelligence (IJCAI-99), pp 1274-1279. Morgan Kaufmann.
  • I Zukerman, D W Albrecht and A E Nicholson (1999). Predicting users' requests on the WWW. In J Kay (ed), CProceedings of the Seventh International Conference on User Modeling (UM-99), pp 275-284.
  • K B Korb, R McConachy and I Zukerman (1997). A cognitive model of argumentation. Proceedings of the Nineteenth Annual Conference of the Cognitive Science Society, 400-405. Lawrence Erlbaum Associates.

• VE Software Tool
  • Software (Tar'd, gzipped).
  • L R Hope, A E Nicholson and K B Korb: Knowledge Engineering Tools for Probability Elicitation, School of Computer Science and Software Engineering, Monash University, 2002, 10pp. Technical report 2002/111. Full TR (postscript), Abstract.

• Matilda: a KEBN software tool
  • Boneh, T (2003). Support for Graphical Modelling in Bayesian Network Knowledge Engineering: a Visual tool for domain experts, Dept. of Computer Science, University of Melbourne, Masters thesis.
  • Software (for Windows, zipped)

Past Projects

• National Australia Bank In 1999 we assisted NAB in the development of Bayesian networks for risk assessment in investment strategies. We trained NAB personnel in the technology and use of Bayesian nets, and assisted them in developing a prototype system.
  

  Participants: Kevin Korb and Ann Nicholson.