Incomplete Knowledge

Our knowledge about the world is necessarily incomplete. In adversarial or competitive situations, the opponent will generally not inform you of all her assets, plans, or capabilities. Even in non-adversarial situations, however, it is never possible to acquire all the relevant information in any but the most artificial domains.

Planning in, or reasoning about, the world given incomplete knowledge thus presents a number of challenges. In some cases, it is possible and sufficient to identify likely states, and reason about them, assuming that mistakes either don't matter or can be corrected as they are discovered. In other situations, it is necessary to explicitly develop solutions that are appropriate for any of the different ways in which things may actually be, thus ensuring that those solutions are robust.

CIRL

Developing robust solutions -- solutions that will work however (perhaps within certain limits) the world turns out -- has recently become a focus of effort at CIRL, largely due to our discoveries about solution structure and ways in which it can be exploited.

We have also been interested in nonmonotonic reasoning, which addresses how to arrive at reasonable default conclusions to fill knowledge gaps in reasonable ways, but lately that interest has been focused mainly on computational, rather than representational, issues.

Pointers

Parent areas:
Modeling
Subareas:
Robust Solutions

Back to the CIRL Overview Page.