DCOP team formation for disaster responsePeople |
|
Research Summary
Human team formation, where one or more teams must be constructed from a large heterogeneous pool of candidates, is one area where DCOP can be applied to practical use. While this can be formulated as centralized COP as well, using k-optimal assignment methods within DCOP is a logical choice to ensure a diverse set of choices, which would be optimal within a group of similar team choices.
In general terms, the formulation considers each person as
an agent with its own domain of tasks or roles it can assume. These agents have
constraints across the task allocations and may have a combination of hard
constraints (which must not be violated), along with soft constraints (which
should be optimized). For example, one hard constraint could be that no two
agents take the same task within a team, while a soft constraint could consist
of two agents performing complementary tasks, which result in a quantifiable
reward.
Distributed Constraint Optimization
Distributed constraint optimization (DCOP) is a framework for cooperative agents, each in control of one or more variables, to work together to optimize a set of constraints that exist upon the variables. This is often visualized as graph structure, with each agent and/or variable as a node, with weighted edges representing constraints. This representation allows k-optimal bounds to give a bounded locally optimal solution.
Relevant Papers and Presentations
|
Document
|
Download
|
|
Local Optimal Solutions for DCOP: New Criteria, Bound, and Algorithm (OptMAS-2009) |