Resilient Monitoring of Social Dynamical Systems through Collaborative Multi-Agent Networks under Latency

Mohammadreza Doostmohammadian; Sergio Pequito

eess.SYAug 25, 2026

Resilient Monitoring of Social Dynamical Systems through Collaborative Multi-Agent Networks under Latency

Authors: Mohammadreza Doostmohammadian, Sergio Pequito

Abstract

Social dynamical networks significantly influence contemporary digital landscapes, affecting realms from social activism to public policy formulation. This paper investigates the use of multi-agent systems (MAS) to monitor and analyze these networks. Firstly, we propose a single-time-scale distributed inference model designed to effectively manage challenges such as latency and agent failure. Secondly, we provide sufficient conditions that ensure the stability of the proposed scheme. Notably, the observer gain design remains effective regardless of time delays. Thirdly, we develop a computationally efficient recovery mechanism for agent failures that relies on employing graph-theoretic approaches to restore network observability by replacing failed agents (due to sensing failure or unbounded delays, i.e., packet drops) by implementing computationally efficient graph-theoretic methods to assign observationally equivalent agent counterparts. Lastly, we illustrate the proposed scheme through a pedagogical example and real-world network applications.

Figures & tables

Symbol	Description
$\mathbf{x}_{k},x_{k}^{i}$	social state (at individual $i$ )
$k$	time index
$A$	social dynamics matrix
$\mathbf{y}_{k},\mathbf{y}_{k}^{i}$	agent observation (at node $i$ )
$H,H_{i}$	observation matrix (at node $i$ )
$\nu_{k},\zeta_{k}$	system and output noise

Table 1: Description of notations and symbols

Figure 1: This figure illustrates a social network monitored by a multi-agent network. The social network has the interactions between individuals depicted by blue directed links, and where some individuals might be just influencers (in red) and others just followers (in blue). Additionally, agents depicted in green harvest data from select individuals enclosed by purple boundaries. These collectors then exchange gathered data through a dedicated communication network, illustrated by green lines.

Figure 2: This figure shows a sample academic social graph $\mathcal{G}_{A}$ (which is not strongly connected) and its parent-child SCC classification. The SCCs with no outgoing links (to other SCCs) are called the parent, and the other remaining SCCs are called the child. The outputs of state nodes in the same parent SCC are observationally equivalent, i.e., the rank of observability matrix given the pair of system matrix $A$ and output $H_{i}$ (of state node $i$ ) are equal for all the nodes in the same parent SCC.

$\overline{\tau}$	0	3	7	15
$\rho(\underline{\widehat{A}})$	0.871	0.970	0.982	0.990

Table 2: The spectral radius of the closed-loop observer for the multi-agent network in Fig. 2 under different bounds on the time-delays.

Figure 3: The time-evolution of the mean-squared error for distributed inference of the social network example in Fig. 2 . The simulation is performed and averaged for $20$ independent Monte-Carlo trials.

$\overline{\tau}$	0	3	7	15
$\rho(\underline{\widehat{A}})$	0.851	0.960	0.980	0.989

Table 3: The spectral radius of the closed-loop observer for the recovered multi-agent network under different bounds on the time-delays.

Figure 4: The time-evolution of the distributed inference mean-squared-error for the recovered multi-agent network. The simulation is performed and averaged for $20$ independent Monte-Carlo trials.

$\overline{\tau}$	0	5	10	20
$\rho(\underline{\widehat{A}})$	0.461	0.879	0.932	0.964

Table 4: The spectral radius of the closed-loop observer for the Sampson’s network example under different bounds on the time-delays

Figure 5: The time-evolution of the mean-squared error for distributed inference of Sampson’s social network. The simulation is performed and averaged for $50$ independent Monte-Carlo trials.

Multi-Agent Systems

Multi-Agent Systems studies how autonomous agents coordinate, compete, communicate, and learn in shared or interdependent tasks.

See all 677 papers

AgentSoV: Virtual Sensing for LLM Multi-Agent Systems via Stream of Variation
Oct 17, 2026Congjing Zhang, Siyuan Xu, Yunfei Bai +3Multi-Agent LLM Systems Multi-Agent Systems
Department of Industrial & Systems Engineering, University of Washington · Department of Electrical & Computer Engineering, Texas A&M University · Amazon
How network perturbations distort agreement trajectories in LTI multi-agent systems
Jul 21, 2026Gal Barkai, Irinel-Constantin MorărescuMulti-Agent Coordination Multi-Agent Control
Université de Lorraine, CNRS, CRAN, F-54000 Nancy, France
How Task Structure Limits Multi-Agent Success: An Information-Theoretic Analysis
Jun 11, 2026Shi Pan, Ming LuoMulti-Agent Collaboration Multi-Agent Systems
University College London London, UK · University of Bristol Bristol, UK

Oct 17, 2026J/K move · Enter open · S save

AgentSoV: Virtual Sensing for LLM Multi-Agent Systems via Stream of Variation

Congjing Zhang, Siyuan Xu, Yunfei Bai, Ke Ren+2

Department of Industrial & Systems Engineering, University of Washington · Department of Electrical & Computer Engineering, Texas A&M University · Amazon

Multi-Agent LLM Systems Multi-Agent Systems Virtual Sensing Failure Attribution for LLM Agents

Open paper AskPDF arXiv

Large language model (LLM) multi-agent systems (MASs) solve complex tasks through interacting agents, yet their reliability is difficult to analyze because failures depend on both individual agents and how information and errors propagate across the system. Existing approaches largely treat this as post-hoc failure attribution, but do not provide a common model across heterogeneous MAS architectures or support broader analysis of information influence and observability. We introduce AgentSoV, a virtual-sensing framework that models MAS execution as a stochastic dynamic system through Stream of Variation (SoV). AgentSoV maps heterogeneous trajectories into diagnostic stages, constructs virtual measurements from observable outputs, and learns cross-stage variation propagation with a state-space model. Its representation is architecture-aware in stage dependencies yet architecture-agnostic in analytical language, supporting chain-structured, orchestration-based, and directed acyclic graph (DAG)-structured MASs. Without re-executing the MAS, AgentSoV supports three system-level analyses: failure attribution identifies where abnormal variation enters; stage-input influence quantifies how strongly intermediate inputs affect downstream outcomes; and virtual-sensor placement selects complementary observations under a monitoring budget. Experiments across multiple MAS architectures demonstrate a unified approach to fault localization, information influence, and observability. AgentSoV reframes agentic execution as a dynamic system to model rather than merely text to inspect, providing insights for MAS structure and component design.

Abstract

Multi-Agent Systems

Explore similar work

AgentSoV: Virtual Sensing for LLM Multi-Agent Systems via Stream of Variation

How network perturbations distort agreement trajectories in LTI multi-agent systems

How Task Structure Limits Multi-Agent Success: An Information-Theoretic Analysis