Modeling

Often, the way the world is modeled (what entities are assumed to exist, and what relationships among them are considered) can be critically important in determining how difficult it is to solve problems.

For example, consider a checkerboard with the single squares at 2 opposite corners removed. Can the board be exactly covered with a set of rectangular tiles each of which covers 2 squares? If the problem is represented naively, discovering the answer depends on searching through all the possible arrangements of tiles. The simple representational device of noticing that the ``mutilated'' board has 32 black squares and 30 white ones, and that each tile covers one square of each colour, however, makes it trivially obvious that the task is impossible.

CIRL

We are exploring ways to simplify the computational problems found in real-world tasks by modeling them appropriately. For example, by breaking planning tasks down hierarchically and restricting information flows between subtasks, it is possible to solve some problems more effectively while simultaneously avoiding solutions that have such undesirable features as high communication overhead.

Modeling is also key in scheduling. For example, by representing the precedence of tasks to be scheduled (what comes before what), rather than the start times for tasks, the size of the search space can be reduced dramatically without loss of generality or functionality.

Pointers

Parent areas:
Knowledge Representation

Subareas:
Agents
Hierarchies
Incomplete Knowledge

Back to the CIRL Overview Page.