David Poole

My goal is to understand the principles behind action in natural and artificial agents. I consider the question: what should an agent do based in its preferences (or goals), its background knowledge, (limited) perception of the environment, previous experience and limited computation?

Currently, my main research interests consider theoretical and empirical investigations of automated reasoning built on the foundations of logic, probability and decision theory for tasks such as diagnosis/recognition, design, planning and reacting with partial knowledge and imperfect sensors. I ask what is a natural compact representation and how can aspects of it be exploited computationally.

My main recent work has been in probabilistic reasoning and assumption-based logical reasoning. This includes developing a coherent theory, and empirical studies based on building efficient implementations, and developing methodologies for various applications. Different behaviour can be characterised by who chooses the assumptions: where nature chooses assumptions we have a logic-based characterisation of recognition/diagnosis incorporating probabilistic models such as Bayesian networks; much of the work on non-monotonic (default) reasoning can be seen as assumption-based reasoning where an adversary chooses the assumptions; where an agent can choose the assumptions we have a characterisation of design and planning. Combining these into a coherent framework that will look something like a logic-based decision/game theory with algorithms that exploit properties of various domains is a medium-term goal of my research.

Other interests include: algorithms for probabilistic reasoning, diagnosis and decision theory; preference elicitation; robotics; user modeling; foundations of diagnosis and recognition; planning under uncertainty and learning.