Distributed Knowledge-based Search
There are two principle ways to use several computers to do knowledge-based search (that also can be applied together):
We are interested in distributed knowledge-based search. In order to understand the problems that have to be solved when trying to come up with a concept for distributed search, let us look at an example for knowledge-based search we are all familiar with: puzzles.
A puzzle consists of many pieces that have one, two or even three sides in common. In order to make the problems in distributed search more obvious, let us also throw in many additional pieces that are not needed to solve the puzzle. The goal of solving a puzzle is to produce a certain picture by putting fitting pieces together.
Solving a puzzle is indead a good example for a knowledge-based search problem:
Now let us assume that a group of people get the task to solve a puzzle. Then they face the following problems:
Each concept for distributed knowledge-based search has to come up with solutions to these problems. We have developed two basic concepts for distributed search that have some features in common but that are aimed at teams of problem solvers that differ in their abilities:
Both concepts have in common that they solve the control problem in the same way: in our puzzle example each of the agents gets its own copy of the puzzle. In fact, both concepts are improvements of the so-called competition approach towards added cooperation of the agents.
A distributed system employing the competition approach just gives the problem to solve to all agents and lets them work on their own until one agent has found a solution.
We also have developed an approach for distributed data mining, called CoLe that combines some concepts from improving on the competition approach with ideas from dividing the problem into subproblems.
Last Change: 5/12/2013