I am now working for Google. However, you can still use this page for access to my publications.

Dr Douglas Aberdeen Senior Researcher Statistical Machine Learning National ICT Australia / ANU email: doug. aberdeen on gmail

The Statistical Machine Learning Program at National ICT Australia is working on intelligent data analysis for bioinformatics, reinforcement learning, document analysis, and learning theory.

I'm in the process of finishing my job with NICTA to start work for Google Zurich. This web page will no longer be maintained, but I'll put in a link from here to a new web page soon. In the meantime, check out my Doug's Rants blog.

Reinforcement Learning (RL)

Allowing automated systems to improve their performance by experience. As long as the system has a set of actions it can do, and a possibly delayed performance signal, we can optimise its behaviour. These systems often (weakly) mimic the way people or animals learn, by being rewarded for good behaviour.

In particular I'm intersted in large scale, real world applications of RL. This involves improving RL algorithms to handle such problems.

Automated Planning

Automated planning constructs complex sequence of primitive actions (or tasks) in order to best achieve the planningn goal. A planning goal might be something like "complete construction of stadium", or "complete all designated experiments on Mars". Primitive actions are things like "pour concrete", or "transmit Mars photo".

I'm particularly interested in planning under uncertainty, where the results of primitive actions can only be predicted with a certain (or uncertain) probability. I am exploring the intersection of RL algorithms and planning.

Cluster Computing

Many of the algorithms I use are amenable to being run on large clusters of cheap computers. This involves some research in how best to split problems up, and merge the results.