VLMDs-Privacy： privacy-enhanced strategy for cooperative decision-making in socially-aware multi-agent systems

doi:10.11772/j.issn.1001-9081.2025050654

Abstract

Abstract:

Socially-aware Multi-Agent Systems （MASs） rely on leadership structures to enhance collaborative decision-making efficiency， but are vulnerable to privacy leakage risks caused by the analysis of state trajectories in Markov Decision Processes （MDPs）， exposing critical nodes to targeted attacks. To address these challenges， a Virtual Leader Minimal Dependency Privacy protection strategy （VLMDs-Privacy） was proposed， which achieved secure and efficient collaborative decision-making in MAS through the following methods： 1） A State Transition Adaptive Differential Privacy mechanism （STADP） was designed to establish a dynamic mapping between state-transition probabilities and privacy budgets， protecting MDP state trajectories from reverse inference attacks； 2） A Virtual Leader Minimal Dependency strategy （VLMDs） was developed to reduce reliance on virtual leaders while achieving globally optimal decision-making， thereby significantly improving resistance to single-point failures； 3） A privacy-efficiency dual-regulation mechanism was constructed to dynamically allocate privacy budgets based on agent behavior credibility， achieving an adaptive trade-off between social awareness and privacy protection. Experimental results showed that under a strong privacy constraint （ $ε$ = 0.1）， VLMDs-Privacy achieved an average arrival success rate of 94.2% in navigation and dynamic maintenance scenarios， outperforming the conventional leader-based differential privacy scheme VLDPs-Privacy （27.9%） by 66.3 percentage points， with only a 3.3% drop compared to non-private settings. These findings validate the robustness of VLMDs-Privacy in maintaining system collaboration capability and privacy preservation efficiency under strong privacy constraints， providing theoretical and technical support for collaborative decision-making in privacy-sensitive MAS deployed in open environments.

Key words: social awareness, Multi-Agent System (MAS), Markov Decision Process (MDP), virtual leader, differential privacy

摘要：

具有社会意识的多智能体系统（MAS）依赖领导结构提升协同决策效率，但易引发基于马尔可夫决策过程（MDP）状态轨迹分析的隐私泄露风险，导致关键节点遭受定向攻击。为此，提出虚拟领导人最小依赖隐私保护策略（VLMDs-Privacy），通过以下方法实现MAS安全高效的协同决策：1）设计状态转移自适应差分隐私机制（STADP），建立状态转移概率与隐私预算的动态映射关系，保护MDP状态轨迹免受逆向推理攻击；2）开发虚拟领导人最小依赖策略（VLMDs），在降低对虚拟领导人依赖的同时实现全局最优决策，显著提升系统抗单点故障能力；3）构建隐私-效率双重调节机制，基于智能体行为可信度动态分配隐私预算，自适应权衡社会意识与隐私保护。实验结果表明，在强隐私约束（ $ε$ =0.1）下，VLMDs-Privacy在导航、动态维修等场景中的平均到达成功率达94.2%，较传统领导人差分隐私方案VLDPs-Privacy （27.9%）提升66.3个百分点，相较于非隐私保护场景仅降低3.3%。该结果验证了VLMDs-Privacy在强隐私约束下维持系统协作能力与隐私保护效能的鲁棒性，可为开放环境中隐私敏感型MAS的协同决策提供理论和技术支撑。

关键词: 社会意识, 多智能体系统, 马尔可夫决策过程, 虚拟领导人, 差分隐私

CLC Number:

TP309.7

Yunle WANG, Xiang FENG, Huiqun YU. VLMDs-Privacy： privacy-enhanced strategy for cooperative decision-making in socially-aware multi-agent systems[J]. Journal of Computer Applications, 2026, 46(5): 1518-1525.

王云乐, 冯翔, 虞慧群. 社会意识多智能体协同决策的隐私增强策略VLMDs-Privacy[J]. 《计算机应用》唯一官方网站, 2026, 46(5): 1518-1525.

Figures/Tables 10

Fig. 1 Decision-making process of socially-aware MAS in making own travel plan

Fig. 2 Flowchart of VLMDs-Privacy

Fig. 3 Comparison of arrival success rates for different algorithms in MAN scenarios

Fig. 4 Comparison of arrival success rates for different algorithms in MADR scenarios

Tab. 1 Arrival success rates of various algorithms under different privacy parameters ε （MANC）

$ε$	到达成功率/%
$ε$	MLD_DP	MDJP_DP	VLDPs-Privacy	VLMDs-Privacy
0.1	91.6	91.6	10.6	94.1
1	92.0	92.1	12.2	94.7
5	92.6	92.5	30.8	94.5
10	94.8	94.9	86.3	95.3
无隐私	96.3	96.2	99.0	96.5

Tab. 1 Arrival success rates of various algorithms under different privacy parameters ε （MANC）

$ε$	到达成功率/%
$ε$	MLD_DP	MDJP_DP	VLDPs-Privacy	VLMDs-Privacy
0.1	91.6	91.6	10.6	94.1
1	92.0	92.1	12.2	94.7
5	92.6	92.5	30.8	94.5
10	94.8	94.9	86.3	95.3
无隐私	96.3	96.2	99.0	96.5

Fig. 5 Arrival success rate for different number of iterations

Fig. 6 Arrival success rate for different privacy parameters

Fig. 7 Influence of global correlation on arrival success rate and iteration convergence

Fig. 8 Arrival success rates of different communities in different initial states （MADR）

Fig. 9 Arrival success rates with only one agent working properly

References 20

[1]	LI Y， MA D， AN Z， et al. V2X-Sim： multi-agent collaborative perception dataset and benchmark for autonomous driving［J］. IEEE Robotics and Automation Letters， 2022， 7（4）： 10914-10921.
[2]	YING C， ZHENG N， WU Y， et al. Privacy-preserving adaptive resilient consensus for multiagent systems under cyberattacks［J］. IEEE Transactions on Industrial Informatics， 2024， 20（2）： 1630-1640.
[3]	LIU Y， ZHOU Y， YANG L， et al. Simulating staff activities in healthcare environments： an empirical multi-agent modeling approach［J］. Journal of Building Engineering， 2024， 84： No.108580.
[4]	KHAN H I， FRIDOVICH-KEIL D. Leadership inference for multi-agent interactions［J］. IEEE Robotics and Automation Letters， 2024， 9（5）： 4671-4678.
[5]	NI J， SHI P. Adaptive neural network fixed-time leader–follower consensus for multiagent systems with constraints and disturbances［J］. IEEE Transactions on Cybernetics， 2021， 51（4）： 1835-1848.
[6]	ZHANG L， LI Y， LU J， et al. Bipartite event-triggered impulsive output consensus for switching multi-agent systems with dynamic leader［J］. Information Sciences， 2022， 612： 414-426.
[7]	WANG P， WEN G， HUANG T， et al. Asymptotical neuro-adaptive consensus of multi-agent systems with a high dimensional leader and directed switching topology［J］. IEEE Transactions on Neural Networks and Learning Systems， 2023， 34（11）： 9149-9160.
[8]	LIU Y， DONG X， SHI P， et al. Distributed fault-tolerant formation tracking control for multiagent systems with multiple leaders and constrained actuators［J］. IEEE Transactions on Cybernetics， 2023， 53（6）： 3738-3747.
[9]	ZHANG H， LI W， ZHANG J， et al. Fully distributed dynamic event-triggered bipartite formation tracking for multiagent systems with multiple nonautonomous leaders［J］. IEEE Transactions on Neural Networks and Learning Systems， 2023， 34（10）： 7453-7466.
[10]	RADFORD A， KIM J W， HALLACY C， et al. Learning transferable visual models from natural language supervision［C］// Proceedings of the 38th International Conference on Machine Learning. New York： JMLR.org， 2021： 8748-8763.
[11]	秦臻，庄添铭，朱国淞，等. 面向人工智能模型的安全攻击和防御策略综述［J］. 计算机研究与发展， 2024， 61（10）： 2627-2648.
	QIN Z， ZHUANG T M， ZHU G S， et al. Survey of security attack and defense strategies for artificial intelligence model［J］. Journal of Computer Research and Development， 2024， 61（10）： 2627-2648.
[12]	YAN B， SHI P， LIM C C， et al. Optimal robust formation control for heterogeneous multi-agent systems based on reinforcement learning［J］. International Journal of Robust and Nonlinear Control， 2022， 32（5）： 2683-2704.
[13]	ZHANG W， JIANG B， LI M， et al. Privacy-preserving aggregate mobility data release： an information-theoretic deep reinforcement learning approach［J］. IEEE Transactions on Information Forensics and Security， 2022， 17： 849-864.
[14]	HASSAN A， DEKA D， DWORKIN Y. Privacy-aware load ensemble control： a linearly-solvable MDP approach［J］. IEEE Transactions on Smart Grid， 2022， 13（1）： 255-267.
[15]	DWORK C. Differential privacy［C］// Proceedings of the 2006 International Colloquium on Automata， Languages， and Programming， LNCS 4052. Berlin： Springer， 2006： 1-12.
[16]	YE D， ZHU T， ZHOU W， et al. Differentially private malicious agent avoidance in multiagent advising learning［J］. IEEE Transactions on Cybernetics， 2020， 50（10）： 4214-4227.
[17]	NGO D D， VIETRI G， WU Z S. Improved regret for differentially private exploration in linear MDP［C］// Proceedings of the 39th International Conference on Machine Learning. New York： JMLR.org， 2022： 16529-16552.
[18]	CHEN B， LEAHY K， JONES A， et al. Differential privacy for symbolic systems with application to Markov chains［J］. Automatica， 2023， 152： No.110908.
[19]	SEO K， YANG J. Differentially private actor and its eligibility trace［J］. Electronics， 2020， 9（9）： No.1486.
[20]	ERDEMIR E， DRAGOTTI P L， GÜNDÜZ D. Privacy-aware time-series data sharing with deep reinforcement learning［J］. IEEE Transactions on Information Forensics and Security， 2021， 16： 389-401.