is a postdoc at Oregon State University working Professor Tom Dietterich. He did his Ph.D. in Computer Science at the University of British Columbia
working with DavidPoole
. He is also pretty much the webmaster for this website. If you have any technical issues with the website or suggestions for changes you can contact him or DavidPoole
. For more up to date information visit Mark Crowley's homepage
Mark can be reached via email on the cs.ubc.ca
domain under the username crowley
To reach Mark as the administrator of this website you can ContactTheAdministrator
Temporal-Spatial Decision Making Under Uncertainty
In many real world domains decisions must be made for many spatial locations over and over into the future. There are also many variables defined over this landscape that taken together with the decisions influence our preference for different states. My research is to find better ways to optimize such decisions when the size of the spatial dimensions are very large and when the variables have many dimensions and change significantly over time dependent loosely on our decisions.
A good example of this comes from Forestry. There is a huge infestation of Mountain Pine Beetles (MPB) in the forests of British Columbia. At its peak, which is expected to be 2006, approximately 10 million hectares (ha) of pine forest will be infested. Once infested the trees inevitably die and their monetary value to the forest industry degrades over two or three years until they are worthless. The social impact is also enormous as anyone driving through central BC can tell you, half the forests are red-brown with the disease. In the short term jobs will be gained trying to salvage as much as possible and much forest cover will be lost, leaving desolate holes. In the long term, many jobs will be lost as less trees are available to be cut. But the forest will also recover, likely with a more diverse species makeup or a totally different one than before.
The problem is what to do about it. Forestry companies, the government and society all have a large stake in influencing this problem. Unfortunately, the scale is so large that only vague predictions of the impact of different strategies is possible at this time. I am looking into ways to model this type of domain more efficiently using hierarchical representations, statistical modelling, reinforcement learning and other AI techniques.
Effect of Conditioning in Belief Networks
In Belief Networks
, conditioning, or setting the state of some variables in the network, has well understood effects on the distributions in the network. There are several interesting effects however that seem to show up, especially in Decision Networks?
that are at first glance non-intuitive. I am exploring how these effects could be usefull for decision modeling.
Non-expected Utility Decision Making
Specifically the impact of theories in Psychology and Economics such as Prospect Theory?
that show ways to properly model people's behaviour in the situations where [Expected Utility Theory] has well known failures?
, Alais' Paradox?
is just one example of such a failure.
You can create latx formatting on the wiki pages by wrapping normal latex code in dollar signs as usual.
does this work?
Why don't these work?
Check the site
Personal Preferences (details in TWikiVariables)
- Show tool-tip topic info on mouse-over of WikiWord links, on or off: (see details in TWikiPreferences)
- Set LINKTOOLTIPINFO = off
- Horizontal size of text edit box:
- Vertical size of text edit box:
- Style of text edit box.
width: 99% for full window width (default),
width: auto to disable.
- Set EDITBOXSTYLE = width: 99%
- Optionally write protect your home page: (set it to your WikiName)