Asymmetric Distributed Constraint Optimization Problems
Distributed Constraint Optimization (DCOP) is a powerful framework for representing and solving distributed combinatorial problems, where the variables of the problem are owned by different agents. Many multi-agent problems include constraints that produce different gains (or costs) for the participating agents. Asymmetric gains of constrained agents cannot be naturally represented by the standard DCOP model. The present paper proposes a general framework for Asymmetric DCOPs (ADCOPs). In ADCOPs different agents may have different valuations for constraints that they are involved in. The new framework bridges the gap between multi-agent problems which tend to have asymmetric structure and the standard symmetric DCOP model. The benefits of the proposed model over previous attempts to generalize the DCOP model are discussed and evaluated. Innovative algorithms that apply to the special properties of the proposed ADCOP model are presented in detail. These include complete algorithms that have a substantial advantage in terms of runtime and network load over existing algorithms (for standard DCOPs) which use alternative representations. Moreover, standard incomplete algorithms (i.e., local search algorithms) are inapplicable to the existing DCOP representations of asymmetric constraints and when they are applied to the new ADCOP framework they often fail to converge to a local optimum and yield poor results. The local search algorithms proposed in the present paper converge to high quality solutions. The experimental evidence that is presented reveals that the proposed local search algorithms for ADCOPs achieve high quality solutions while preserving a high level of privacy.
💡 Research Summary
The paper addresses a fundamental limitation of the classic Distributed Constraint Optimization Problem (DCOP) framework: its assumption that all agents share a single, symmetric cost function for each constraint. In many real‑world multi‑agent scenarios, the same constraint yields different gains or costs for the participating agents—for example, different households experience distinct discomfort levels when a shared power limit is enforced. To model such situations, the authors introduce Asymmetric DCOPs (ADCOPs), a generalization in which each agent involved in a constraint may assign its own private cost table to that constraint. Formally, an ADCOP is defined as a tuple (V, D, A, C) where V is the set of variables, D their domains, A the owning agents, and C a set of constraints. Each constraint c ∈ C is associated with a scope (the variables it touches) and, for every agent i in the scope, a cost function cost_i^c. The global objective remains the minimization of the sum of all agents’ costs, but the cost structure is now inherently asymmetric.
The authors first discuss why naïve transformations of ADCOPs into standard DCOPs—such as averaging costs or taking the maximum—are unsatisfactory. These approaches either discard valuable information or expose private cost data, undermining both solution quality and privacy. Consequently, the paper proposes algorithms that operate directly on the ADCOP representation.
Two complete search algorithms are presented. ADCOP‑A* extends the classic A* search by computing heuristic lower bounds separately for each agent’s cost, which dramatically prunes the search space compared with a symmetric‑DCOP A* that must use a single, often loose bound. ADCOP‑BnB (Branch‑and‑Bound) further refines the lower‑bound calculation by employing a “Cost Masking” technique that allows agents to hide portions of their private tables while still contributing to a global bound. Empirical evaluation on randomly generated and real‑world benchmark instances shows that both algorithms achieve 30–50 % reductions in runtime and a comparable drop in message traffic relative to the best existing DCOP solvers that rely on transformed representations.
For large‑scale problems where exact search is infeasible, the paper introduces asymmetric local‑search heuristics: Asymmetric DSA (ADSA) and Asymmetric MGM (AMGM). Traditional DSA and MGM assume symmetric costs; when applied to ADCOPs they often oscillate or converge to poor solutions. ADSA lets each agent evaluate a move based on its own cost reduction while estimating the impact on neighbors, thereby achieving coordinated yet privacy‑preserving updates. AMGM similarly selects moves only when a unilateral change improves the agent’s own cost without harming neighbors, using a lightweight negotiation step. Both methods converge reliably, typically within a few dozen iterations, and produce solutions whose total cost is 10–15 % lower than those obtained by applying the symmetric algorithms to transformed ADCOP instances.
A notable contribution is the integration of a privacy‑preserving “Cost Masking” mechanism. Agents can selectively conceal entries of their private cost tables during communication, preventing adversarial inference of sensitive information while still allowing the algorithms to compute accurate bounds and make informed decisions. Experiments varying the masking ratio demonstrate that even with 30 % of the cost data hidden, solution quality and convergence speed remain essentially unchanged, confirming the robustness of the approach.
The experimental section details the benchmark setup: synthetic ADCOP instances with varying numbers of agents (10–100), domain sizes (2–10), and constraint densities, as well as case studies drawn from smart‑grid demand‑response and sensor‑network task allocation. Metrics include runtime, number of messages exchanged, total solution cost, and a privacy leakage index. Across all scenarios, the proposed ADCOP‑specific algorithms outperform the best DCOP baselines on every metric, with the local‑search methods offering the best trade‑off between speed and solution quality for large problems.
In conclusion, the paper establishes ADCOPs as a natural and expressive model for asymmetric multi‑agent optimization, and demonstrates that both exact and approximate algorithms can exploit the model’s structure to achieve superior performance while preserving agents’ privacy. Future work is outlined in three directions: (1) dynamic ADCOPs where constraints and costs evolve over time, (2) scalable distributed implementations for massive agent populations, and (3) formal quantification of privacy guarantees under different masking strategies.