I am currently a Postdoctoral scholar in Computer Science at Oregon State University working with Tom Dietterich. My primary research is into methods for decision making under uncertainty in massive spatiotemporal domains such as forest fire management and invasive species control. Other keywords that relate to my research interests within Artificial Intelligence include probabilistic graphical models, probabilistic inference, machine learning and reinforcement learning.
The field of Computational Sustainability is at the intersection between computational sciences (such as artificial intelligence, computational modelling, optimization and planning research) with applied research in environmental/ecological domains (such as land use management, invasive species spread, sustainable ecology management, smart grids and species tracking).
Some of the major problems I am looking at are:
- How can we compactly represent expressive and interpretable policies for acting in large spatial domains with correlated actions?
- How can we effectively and efficiently find optimal (or approximately optimal) policies for large spatiotemporal problems?
- If our optimization approach is approximate, how do we bound the error so that we know something about how close we are to optimal?
- What general approaches can we devise for reasoning about spatial planning problems?
Major Contributions and Interests
- Inference in Probabilistic Graphical Models : Modelling constraints in Bayesian Networks using conditioned nodes. Unexpected biases can arise in the joint distribution if this is done naively without considering the way inference is performed. I showed a way to shield part of the BN from this bias while maintaining the constraint. (M.Sc. Thesis 2005, CAI2007)
- Reinforcement Learning : My recent research falls under the general category of Reinforcement Learning. There is a lot of exciting research happening in this field lately with the growing power of PAC-MDP algorithms.
- I proposed using equilibrium policies as a representation for policies in large spatiotemporal planning problems and demonstrated how they could be used to perform policy gradient planning. (AAAI2011, Ph.D. Thesis 2011)
- I used forestry planning as my demonstration problem. I looked at the management of forest ecosystems under spatial constraints and under the presence of stochastic spatial disturbances moving across the landscape such as Mountain Pine Beetles. (UAI2009, Ph.D. Thesis)
- Spatial Policy Visualization : how do you compactly represent all of the options and features that are contained in a policy for acting across space and time? What types of visualization are useful for real world practitioners? How could existing planning algorithms need to be modified to take advantage of interactive feeback from users based on a visualization?
- There is a wide range of research in AI, ML, OR and Economics which can be brought to bear on these problems and one of the challenges is bridging the different languages and terminology of different research communities working on similar problems. I believe that understanding and leveraging computation is essential to finding ways for human society to live sustainably on the Earth in the long term. Looking at the planning problems that arise in the real world and then exploring appropriate algorithms offers a fresh approach to advancing computer science and OR research by forcing us to get outside out usual boxes of toy problems and push our methods to their limits.
- I’ve made an informal map of some optimization methods which are suitable for different scales of spatiotemporal problems. I’m constantly adding to it, any feedback is welcome.
- Inference first order lifted probabilistic models
Practical use of graphical models to aid human decision making. In particular, the pitfalls people can fall into if they don’t understand the models or how inference is performed.
- Inference methods in graphical models
- Modelling and inference of causal interactions
- Learning causal models
- Game Theory, Mechanism Design, Behavioural Economics
- Preference elicitation methods, voting systems
- see my existing page at UBC here for more details on my other interests