Community Research and Development Information Service - CORDIS

Making decisions in an uncertain environment

An EU-funded project has developed efficient computational methods for automated learning and decision making under uncertainty. These techniques can be used in many applications such as network security, robotics and traffic management.
Making decisions in an uncertain environment
Statistical decision theory offers a framework for finding optimal solutions but, in many cases, exact inference and planning are intractable. Helping to overcome these obstacles was the aim of the 'Efficient sequential decision making under uncertainty' (ESDEMUU) project, funded by the EU. The project sought to develop efficient approximate methods for nearly optimal learning and decision making in such problems.

One interesting puzzle concerns reinforcement learning in which the intelligent or learning agent must learn to act in an unknown environment, solely via interaction, to maximise the total reward obtained during its lifetime. However, one particular difficulty arises when the parameter distribution is hard to compute. For that reason, the project developed models that can be used with distributed value functions so as to perform decision making in these specific learning problems.

Another sequential decision-making problem involves learning how to act from demonstrations. Here, the project extended a principled generalisation to the problem of learning from multiple teachers with different preferences.

In the distributed constraint optimization problem (DCOP) system, ESDEMUU solved the problem of finding an optimal policy for a group of communicating agents with only limited communication. It developed algorithms that can solve such problems efficiently.

Finally, the now completed project proposed a novel framework to model sparse reward processes, where a learning agent is placed in an unknown environment and is faced with a series of opposing goals for selection. The main use for this approach will be information and communication technologies (ICT) applications such as security and network-intrusion detection.

Related information

Follow us on: RSS Facebook Twitter YouTube Managed by the EU Publications Office Top