TY - JOUR
T1 - Differentially Private Distributed Resource Allocation via Gradient Perturbation
AU - Li, Qingke
AU - Zhou, Jialing
AU - Lv, Yuezu
AU - Wen, Guanghui
N1 - Publisher Copyright:
© 1963-2012 IEEE.
PY - 2026
Y1 - 2026
N2 - Despite progress in distributed resource allocation (DRA) over multi-agent networks, it remains challenging to make allocation decisions while preserving sensitive information during inter-agent communication, particularly over general directed graphs. The hard balance constraint, which requires supply-demand consistency throughout the execution process, further complicates the problem. To address these challenges, this pa per proposes a gradient-perturbed framework for differentially private distributed resource allocation that ensures ε-differential privacy, almost sure convergence, and hard balance constraint satisfaction. For general directed networks, the framework employs an affine transformation based on the out-degree Laplacian matrix, along with gradient estimation and a fixed step size, ensuring almost sure convergence to the optimal solution at a linear rate. A reduced-order variant is also developed for undirected graphs to enhance communication and computational efficiency by exploiting network symmetry. Theoretical analysis establishes the convergence and privacy guarantees, and numerical simulations confirm the effectiveness of the proposed approach.
AB - Despite progress in distributed resource allocation (DRA) over multi-agent networks, it remains challenging to make allocation decisions while preserving sensitive information during inter-agent communication, particularly over general directed graphs. The hard balance constraint, which requires supply-demand consistency throughout the execution process, further complicates the problem. To address these challenges, this pa per proposes a gradient-perturbed framework for differentially private distributed resource allocation that ensures ε-differential privacy, almost sure convergence, and hard balance constraint satisfaction. For general directed networks, the framework employs an affine transformation based on the out-degree Laplacian matrix, along with gradient estimation and a fixed step size, ensuring almost sure convergence to the optimal solution at a linear rate. A reduced-order variant is also developed for undirected graphs to enhance communication and computational efficiency by exploiting network symmetry. Theoretical analysis establishes the convergence and privacy guarantees, and numerical simulations confirm the effectiveness of the proposed approach.
KW - Distributed resource allocation
KW - differential privacy
KW - directed graphs
KW - multi-agent networks
UR - https://www.scopus.com/pages/publications/105053437162
U2 - 10.1109/TAC.2026.3741822
DO - 10.1109/TAC.2026.3741822
M3 - Article
AN - SCOPUS:105053437162
SN - 0018-9286
JO - IEEE Transactions on Automatic Control
JF - IEEE Transactions on Automatic Control
ER -