Mark's Research

My primary interests at the moment are decision making under uncertainty for large spatial-temporal domains by I have a wide interest in many AI and CS topics. Take a look at my If you are interested in similar things and are looking for something to read you can look at my writing or my online bibliography library. Please feel free to contact me if you are interested in similar topics and would like to discuss it.

Spatial-Temporal Decision Making (PhD Thesis)

My research focussed on large scale spatial-temporal decision making under uncertainty. The motivating problem is forestry planning in the presence of dynamic insect infestations such as the Mountain Pine Beetle (MPB) in the BC forests. A lot of work has been done in forestry to learn the behaviour of the beetle and test different strategies. Simulations exist that predict its growth and dispersal patterns. From an AI point of view the next level should be an automated planning system that finds an optimal forest management policy that maximizes the user's utilities for different states of the forest. I am working on ways to find these optimal policies for the very large, high dimensional domains that are being dealt with here. New techniques of abstraction and spatial reasoning will need to be integrated with existing statistical policy search techniques. The goal is to produce an algorithm that takes as input the user's values, the current state of the world and connections to domain specific simulations of the world. The output will be a policy consistent with the user's values. This will allow all interested stakeholders from forestry, the government and society at large to debate the effect of different values on policy and set of values is the most appropriate one.

Using Conditioning to make Constraints (MSc)

My master's thesis work was concerned with the effects of observations in a Bayesian network. It is an important feature of Bayes nets that conditioning, setting the state of a node, influences the posterior probabilities of ancestral nodes in the network. My thesis was that in cases where the conditioned node is not an observation but is rather a constraint, such as 'or', then this propagation of influence is usually not desired. I show examples of how this kind of conditioning side-effect leads to variables being incorrectly tied together in the network. I devised a method of augmenting a Bayes net , called shielding, so that these side-effects are exactly cancelled out. This is done by adding another set of nodes mimicking part of the original nets structure and computing their conditional probability tables. I showed that in many cases inference in this shielded network will be no more complex than in the orignal network. The only extra cost is the original precompilation of the tables for the new nodes which in my work was done using constrained nonlinear optimization. I recently presented a paper on this work at the 20th Canadian Conference on Artificial Intelligence in Montreal.

Game Theory

I am very interested in several topics in Game Theory and regularly attend Kevin Leyton-Brown's reading group GT-DT. My primary intrests in this area are Election Theory, Preference Elicition, Bounded Rationality and Mechanism Design.

First Order Probabilistic Logic

I regularly attend David Poole's reading group FOPI. I am interested in the development of first order languages that fully integrate uncertainty. Particularly, the importance of doing inference and reasoning while remaining in the first-order domain in order to avoid the curse of dimensionality incurred by grounding out to explicit knowledge. I'm also very interested in hierarchical representations as a means of modelling different levels of abstraction. Interesting work in my group has been done on this by Rita Sharma.

Collborative Community Tools (UAIWiki)

I maintain the website for the Association for Uncertainty in Artificial Intelligence (AUAI). In 2005 I added the UAIWiki. A wiki allows many people to edit the same webpage. The content of a wiki is determined by the overall community. All users are equal and can correct each other. It is the process of debate writ onto the web. Debate is the heart of science. So I am hopeful that the UAIWiki will prove to be useful for the UAI community.