Enforcing Soft Local Consistency on Multiple Representations for DCOP Solving

Connecting soft arc consistency with distributed search in DCOP solving has been very beneficial for performance. However, including higher levels of soft arc consistency breaks usual privacy requirements. To avoid this issue, we propose to keep different representations of the same problem on each agent, on which soft arc consistencies are enforced respecting privacy. Deletions caused in one representation can be legally propagated to others. Experimentally, this causes significant benefits.