|
|
|
|
|
|
Help | Seminars List | Add Seminar | Edit Seminars | Tips for organisers | RSS | ics Calendar | Search | Send comments about this website to seminar-master@cecs.anu.edu.au
Contact: luke.fletcher@anu.edu.au INFOENG SEMINAR SERIES Colloquium series
Building Probabilistic Temporal Planning ToolsDr. Douglas Aberdeen (Statistical Machine Learning, National ICT Australia)DATE: 2006-01-13 TIME: 11:00:00 - 12:00:00 LOCATION: RSISE Seminar Room, ground floor, building 115, cnr. North and Daley Roads, ANU ABSTRACT: Probabilistic temporal planning is all about the kind of reasoning humans do all the time. We reason about the best actions to take, like driving a car versus riding the bus, in the midst of the uncertainty of every day life. Sometimes we reason about things to do in parallel. You can take the kids to school while your partner does the shopping. Humans are good at this but can become overwhelmed by complexity, and we do not deal with probabilities precisely. The long- term goal of our research is to produce user friendly tools that reason about the best course of action in complex and uncertain environments. To date, the work of the Project has produced 3 out of the 5 planners that can cope, to some extent, with the complexity of these domains. This talk will discuss how to formalise these problems as Markov decision processes, and present contrasting approaches to solving these very large MDPs. The first method I discuss is based on a form of dynamic programming designed to scale to large state spaces. The second is a numerical optimisation approach that scales to even larger state spaces, at the cost of significant approximations. BIO: Doug's background in Reinforcement Learning, which studies how computers can learn through punishment and reward. He now applies this to temporal planning under uncertainty, which might also be known as "how to build a better project management tools." Doug also has a somewhat historical interest in cluster computing. Doug is contributing to the Dynamic Planning, Learning and Optimisation Project (DPOLP), GymAware and Angie NICTA Projects. Doug completed his Ph.D in Engineering in 2003. He studied at Australian National University, including 1 year at Carnegie Mellon University. He also has a B.E (Electrical Engineering) and B.Sc (Computer Science) from Melbourne University. Doug is also a member of the IEEE.
|