This manuscript introduces a novel preassigned-time distributed optimization protocol for resource allocation in multiagent systems, addressing both local convex set constraints and global equality constraints. The protocol operates through two sequential phases: initially, each agent’s state is deterministically driven into its feasible set within a prespecified time; subsequently, global equality constraints are continuously maintained while progressively converging to the optimal solution, achieving exact optimization within the total preassigned time. Distinct from conventional finite-time, fixed-time, or predefined-time distributed algorithms, our framework innovatively employs a state-based generator mechanism. A key advantage is the settling time’s invariance to both initial conditions and system parameters, enabling precise offline determination of convergence timelines and offering substantial practical benefits for real-time implementations. Numerical experiments validate the theoretical soundness and practical efficacy of the proposed methodology in resource-constrained distributed coordination scenarios.

Federated Learning (FL) enables decentralized model training while preserving data privacy, but it remains susceptible to backdoor attacks, especially under dynamic client participation and non-IID data. Existing attacks, such as Distributed Backdoor Attack (DBA) and Shadow Attack, often struggle to balance stealthiness, robustness, and persistence. To address these challenges, we propose the Stealthy Associate Trigger Attack (SATA), which integrates four key techniques: trigger splitting, associative learning, L2-norm control, and progressive backdoor injection. Experimental results on CIFAR-10 using ResNet-18 demonstrate that SATA achieves a 99.26% Attack Success Rate (ASR) with only 10% malicious clients, while maintaining L2-norm deviations within 0.1 to evade detection. Moreover, SATA achieves an ASR-100 of 86.68%, indicating strong persistence. Additional evaluations on MNIST and under defense mechanisms like Multi-Krum and FoolsGold validate the generalizability and adaptability of our method. These findings highlight critical vulnerabilities in privacy-sensitive FL systems and call for the development of more effective defenses. Future work will explore extending SATA to text and audio domains and improving efficiency on edge devices.

The proliferation of malware poses critical threats to distributed service computing architectures like cloud-edge systems supporting IoT and mobile services. While federated learning (FL) offers privacy-preserving collaborative malware detection by training models locally on distributed clients, its decentralized nature introduces severe vulnerabilities to adversarial backdoor attacks. This study reveals how attackers exploiting FL systems can implant hidden triggers causing targeted misclassification of malware as benign. We propose the Federated Adaptive Backdoor Attack (FABA), which optimizes trigger generation via gradient matching and parameter conformity constraints to ensure stealth while maintaining model performance. Evaluated across Virus-MNIST and MalImg datasets in realistic FL deployments, FABA achieves 100% attack success rates with just 3% malicious clients and minimal triggers (9 pixels), evading detection by state-of-the-art defenses like Multi-Krum and RFA. Crucially, FABA induces negligible degradation in clean-sample accuracy ( < 0.5%) and service latency overhead, demonstrating its feasibility in operational settings. This work exposes critical security risks in privacy-preserving service architectures, providing actionable insights for developing robust FL-based detection systems. Our findings establish essential benchmarks for securing distributed learning in production environments where model integrity directly impacts service reliability.

Federated learning allows multiple clients to locally train on private datasets, followed by the central server aggregating these parameters to form a global model. This process maintains the privacy of client data while enabling the global model to learn from diverse data distributions, thus enhancing its generalization capability. However, it cannot be ensured that all clients participating in the federated aggregation process are fully trustworthy. Some clients might maliciously alter the parameters they upload to the central server, embedding backdoors into the global model. This causes the model to behave in ways dictated by attackers for specific inputs. Existing backdoor defense mechanisms primarily operate on the central server side, attempting to eliminate malicious clients by analyzing parameter characteristics or behaviors, or by pruning or distilling the aggregated model. In this paper, we first reveal the impact of backdoor attacks on neural networks at the neuron level and analyze the parameter changes during backdoor training from the attacker’s perspective. Building on this analysis, we propose a neural network honeypot mechanism. This approach initially uses parameter perturbation and masking for adversarial training to identify fundamental malicious neurons. These neurons are then processed to lure attackers into traps, thereby enhancing the accuracy of malicious neuron identification. We conducted extensive experiments under different adversarial settings and assumptions, demonstrating that our backdoor defense approach consistently surpasses current state-of-the-art baseline methods.

The main aim of this work is to design a non-fragile sampled data control (NFSDC) scheme for the asymptotic synchronization criteria for interconnected coupled circuit systems (multi-agent systems, MASs). NFSDC is used to conduct synchronization analysis of the considered MASs in the presence of time-varying delays. By constructing suitable Lyapunov functions, sufficient conditions are derived in terms of linear matrix inequalities (LMIs) to ensure synchronization between the MAS leader and follower systems. Finally, two numerical examples are given to show the effectiveness of the proposed control scheme and less conservation of the proposed Lyapunov functions.

This paper is concerned with inertial bidirectional associative memory neural networks with mixed delays and impulsive effects. New and practical conditions are given to study the existence, uniqueness, and global exponential stability of anti-periodic solutions for the suggested system. We use differential inequality techniques to prove our main results. Finally, we give an illustrative example to demonstrate the effectiveness of our new results.

The semi-global output consensus problem for multiagent systems depicted by discrete-time dynamics subject to external disturbances and input saturation over switching networks is investigated in this paper. Assume that only a small part of subsystems have directly received the output of the exosystem. The distributed consensus algorithms are proposed by adopting the low-gain state feedback and the modified algebraic Riccati equation. Then, the outputs of all subsystems can reach synchronization asymptotically with those of the exosystem by using the proposed consensus protocols on some preconditions. Both the connected switching networks and the jointly connected switching networks are considered for the semi-global output consensus problem, respectively. Some numerical simulation results are shown to validate the theoretical analysis.

The reduced-order filtering problems are investigated in this paper for static neural networks with Markovian jumping parameters and mode-dependent time-varying delays. By fully making use of integral inequalities, the designs of reduced-order and filters are discussed. The proper gain matrices of filters and the optimal performance indices are efficiently obtained by resolving corresponding convex optimization problems with the constraints of linear matrix inequalities. It is verified that the computational complexity for the reduced-order filter design is significantly reduced when compared with the full-order one. Furthermore, the nonfragile reduced-order filtering problems are also resolved in this paper. Two examples with simulation results are presented to demonstrate the feasibility and application of the established results.

The problem of leader-following consensus of nonlinear agents in hybrid varying directed topology is considering not only the agent but also that the directed edges can have a time-varying nonlinear dynamics with jump discontinuity, which contains the switching topology as its special case. This paper has the following contributions toward this problem. The leader-following consensus problem in hybrid varying directed topology is first addressed, and an online leader switching method is proposed, which reduces the dependence on some global conditions and the connectivity assumptions on the selection of leaders. Second, we generalize the Lipschitz condition and the combined condition of one-sided Lipschitz and quadratically inner-boundedness conditions to a new generalized linear incremental condition, which gives us a more generalized result in the Lyapunov proof and better performance in simulation. Third, an observer-based consensus protocol is constructed with two sufficient observability and controllability conditions and two optimal control design algorithms. Finally, an example of teleoperating multirobotic manipulator joint network is provided to illustrate the performance improvement by comparing with the existing results.