Decentralized Deep Learning Using Momentum-Accelerated Consensus

On global leader-following consensus of identical linear dynamic systems subject to actuator saturation

This paper investigates the distributed consensus problem of fractional-order multi-agent systems under a time-invariant communication topology, where the dynamics of each agent is described by a general fractional-order differential equation. To achieve consensus, a fractional-order observer-type consensus protocol based on relative output measurements is introduced. By using tools from Lyapunov stability theory for fractional-order systems, two theorems about the consensus of fractional-order multi-agent system with a fixed communication topology having a spanning tree are then proposed. Finally, the effectiveness of the theoretical results is demonstrated through numerical simulations.

Wireless networked control systems (WNCS) over mobile ad hoc network (MANET) is a new area of research and has many potential applications, for instance, military or rescue missions, exploring hazardous environments, and so on. For performance evaluation, researchers mostly rely on computer simulations as WNCS experiments are expensive to execute. It will generate a significant benefit to conduct performance analysis of WNCS over MANET using co-simulation that utilises SIMULINK and OPtimised Network Engineering Tool (OPNET) to simulate plant/controller behaviour and the MANET, respectively. Previous conference papers by the authors reported the initial SIMULINK-OPNET co-simulation for only one network size. Here the authors present an extention of their previous work, and presents the SIMULINK-OPNET co-simulation, methodology and comprehensive simulation results that have not been reported previously. It also considers the impact of five network sizes with stationary and mobile nodes. The proposed SIMULINK-OPNET co-simulation is applied to WNCS over MANET using a realistic wireless communication model. It investigates the impact of network data rates, node mobility, the packet delay, packet drop on the system stability and performance.

ABSTRACTBrain tumors pose a significant threat to human health, necessitating early detection and accurate diagnosis to enhance treatment outcomes. However, centralized data collection and processing encounter challenges related to privacy breaches and data integration due to the sensitivity and diversity of brain tumor patient data. In response, this paper proposes an innovative federated learning‐based approach for brain tumor detection, facilitating multicenter data sharing while safeguarding individual data privacy. Our proposed federated learning architecture features each medical center as a participant, with each retaining local data and engaging in secure communication with a central server. Within this federated migration learning framework, each medical center independently trains a base model on its local data and transmits a fraction of the model's parameters to the central server. The central server leverages these parameters for model aggregation and knowledge sharing, facilitating the exchange and migration of models among participating medical centers. This collaborative approach empowers individual medical centers to share knowledge and experiences, thereby enhancing the performance and accuracy of the brain tumor detection model. To validate our federated learning model, we conduct comprehensive evaluations using an independent test dataset, comparing its performance with traditional centralized learning approaches. The experimental results underscore the superiority of the federated learning‐based brain tumor detection approach, achieving heightened detection performance compared with traditional methods while meticulously preserving data privacy. In conclusion, our study presents an innovative solution for effective data collaboration and privacy protection in the realm of brain tumor detection, holding promising clinical applications. The federated learning approach not only advances detection accuracy but also establishes a secure and privacy‐preserving foundation for collaborative research in medical imaging.

Electroencephalography (EEG)-based brain cognitive state identification for the elderly allows timely detection and early intervention of cognitive deterioration. Notably, EEG signals carry a great deal of vital personal information. However, a majority of the existing cognitive evaluations focus on improving the accuracy of EEG decoding and enhancing the performance of identification models, while neglecting the privacy protection of EEG data. To address the risky challenge, we propose a privacy-preserving personalized distributed learning framework (P3DL) for cognitive state identification. Specifically, it consists of the clients and a central server. Each client contains a cognitive model and a score model for identifying cognitive states and quantifying cognitive levels, respectively. The central server can aggregate local models' parameters from distributed clients, then, update and downstream the global model's parameters for iterative optimization. A federated dynamic update strategy (FedDBS) is designed to jointly update all global and local models with a supervisory metric. In order to further improve the identification performance and judge the misdiagnosis level, a novel loss function, extreme error Loss (E2Loss), is proposed. Compared with the baseline, experimental results on our self-collected clinical dataset and a public dataset show an average increase in F2Score of 5.58% and 3.31%, and in accuracy of 1.78% and 2.46%, respectively. Furthermore, the scalability of the framework has been proved in the emotion recognition task. Our proposed framework P3DL can not only improve the identification performance, but also protect the privacy of EEG, opening a new window for secure healthcare.

The accelerating complexity of cyberattacks has pushed conventional Intrusion Detection Systems (IDS) to the brink, exposing critical weaknesses in centralized architectures. Modern attack vectors exploit structural fragilities by tampering with logs, manipulating alert records, and creating blind spots that undermine forensic accuracy. In response to these systemic challenges, the fusion of blockchain technology with intrusion detection has emerged as a promising paradigm for building resilient, transparent, and tamper-proof security infrastructures. This survey consolidates insights from recent advances in blockchain-enabled IDS models to understand how decentralized principles can transform threat monitoring and response. 22 contemporary frameworks, the study evaluates design choices related to data immutability, consensus protocols, threat intelligence sharing, and real-time anomaly detection. Across the literature, blockchain’s core strengths, distributed storage, cryptographic integrity, and verifiable audit trails consistently address the limitations of traditional IDS architectures. However, persistent barriers remain, particularly in scaling blockchain networks for high-volume traffic, managing latency in consensus processes, and ensuring seamless interoperability with heterogeneous network environments. To bridge these gaps, we propose a hybrid IDS architecture that couples federated anomaly detection with decentralized threat validation. The model leverages federated learning to analyze network behavior collaboratively across multiple nodes without exposing raw data. This distributed intelligence layer reduces the training overhead on central servers while improving anomaly recognition across diverse environments. Detected events are then logged onto a lightweight blockchain optimized for low-latency operations. Smart contracts automatically validate alerts based on predefined security conditions, ensuring that only verified, high-confidence events are appended to the ledger. Preliminary evaluation suggests that such a dual-layer approach can significantly enhance system reliability. By distributing both detection and verification tasks, the framework reduces opportunities for log manipulation almost entirely. Meanwhile, smart contract governance provides 90–95% assurance of log integrity, creating an auditable, tamper-resistant record of all intrusion-related activity.The accompanying project extends these principles into a practical implementation. In this system, blockchain functions as the backbone for secure alert management, offering verifiable and persistent storage without reliance on a centralized authority. The machine learning engine processes real-time network traffic, identifying anomalies and classifying threats with precision. Automation through smart contracts streamlines response workflows by handling validation and triggering relevant countermeasures, reducing human error and ensuring rapid containment. Altogether, the proposed Blockchain-Enabled Intrusion Detection System demonstrates a forward-looking security architecture capable of withstanding modern cyber threats. It reinforces organizational trust by eliminating opaque logging practices, reduces manipulation risks inherent in centralized IDS deployments, and enables transparent threat intelligence exchange across distributed stakeholders. By aligning decentralized trust mechanisms with intelligent anomaly detection, the framework marks a significant step toward secure, scalable, and accountable cybersecurity ecosystems. Detection accuracy improves through collective learning, with estimated gains of 30–40% compared to isolated IDS setups

Mobile Ad Hoc Network (MANET) is collection of self configuring, wireless, multi hop mobile nodes. This network lacks centralized routing capability as well as a fixed network infrastructure and topology. In order to communicate with each other these multi hop nodes themselves act as routers. As the nodes are mobile and due to lack of central server, the routing may become erroneous. Also due to dynamic nature of nodes the routing becomes a highly tedious task. In the literature several routing protocols have been proposed for mobile ad hoc networks and most prominent among them are DSR, AODV and TORA. But almost all of these protocols assume that the nodes in the MANET are always cooperative. However, these protocols protect MANET from all the possible attacks except non cooperative behavior. ARAN (Authenticated Routing for Adhoc Networks) is a solution proposed which provides security against all possible vulnerabilities but an authenticated selfish node can manipulate these protocols and can disturb the network by dropping packets or delaying the packet. These noncooperative nodes may in turn partition the network. In this paper we propose a credit based system that can be applied to routing protocol to detect selfish node and improve the network performance.

Vehicular communication networks are a type of network in which vehicles and roadside units are the communicating nodes, providing each other with information such as safety warnings and traffic information. As a cooperative approach, vehicular communications can be more effective in avoiding accidents and traffic congestion, rather than each vehicle trying to solve these problems individually. Therefore, intelligent transportation systems (ITS) have been developed to address the challenges of the safety, security, and efficacy of the current transportation systems. The field of inter vehicular communications (IVC) includes both vehicle-to-vehicle (V2V) and vehicle-to infrastructures (V2I) communication and is also known as the vehicular ad hoc network (VANET). The VANET is recognized as an important component of ITS in various national plans [1]. Direct communication between vehicles may be set up by means of mobile ad hoc networks (MANETs), which do not depend on the altered foundation. Research on MANETs covers the application prerequisites and correspondence conventions for everything from sensor systems to handheld PCs and vehicular frameworks [2].

Mobile Ad hoc Networks (MANET) are wireless infrastructure less networks that are formed spontaneously and are highly dynamic in nature. Clustering is done in MANETs to address issues related to scalability, heterogeneity and to reduce network overhead. In clustering the entire network is divided into clusters or groups with one Cluster Head (CH) per cluster. The process of CH selection and route optimization is extremely crucial in clustering. Genetic Algorithm (GA) can be implemented to optimize the process of clustering in MANETs. GA is the most recently used advanced bio-inspired optimization technique which implements techniques of genetics like selection, crossover, mutation etc. to find out an improved solution to a problem similar to the next generation that inherits the positive traits and features of the previous one. In this chapter several genetic algorithm based optimization techniques for clustering has been discussed. A comparative analysis of the different approaches has also been presented. This chapter concludes with future research directions in this domain.

With the recent advancements in the field of wireless networks or Mobile Ad-hoc networks (MANETs), mobile computing is the most powerful use for network communication and connectivity. Effective networks come across with numerous difficulties. Networks need to be able to transmit data with acceptable precision from one system to another. A framework must ensure that the retrieved data is consistent with the transmitted data for most applications. If the frame between the two nodes is distorted in the data-link layer, it must be corrected before having driven to other nodes. However, most protocols with the link-layer simply dismiss the frame and allow the high-layer protocols to pass down the frame. In other words, information is a valuable commodity, since asset information must be protected from threats. Some applications require a network mechanism for detecting and preventing these attacks in MANETs. An important challenge in MANET is building the secure intrusion detection system in network providing security to the nodes and route paths in network. The attacks in network can threaten the security issues which have been identified in the intrusion detection system engine, later it is prevented by intrusion prevention engine in the network. A henceforth new technique to implement the security goals and prevent attacks is implemented by introducing the Secure-Intrusion Detection System (S-IDS) in the network. The research work introduced the Secure Energy Routing (SER) protocol for the MANETs. The protocol solves the issue of security in network by detecting the attacks and preventing them in the network. The simulation results show the higher packet delivery ratio and low end-to-end delay with and without attacks. The protocol performance is good in terms of its packet delivery ratio and low end-to-end delay respectively.

Economic Power Dispatch is a critical component of the power system network (EPD). It assigns required load to online producing units while accounting for system constraints. The fundamental purpose of EPD is to lower overall power generation costs. For example, the bi-objective Combined Emission- Economic Dispatch (CE-ED) problem addresses the environmental impacts of fossil fuel power plant emissions. The bi-objective challenge reduces both fuel costs and emissions. Generators are simulated using quadratic cost functions to match their output restrictions. The non-linear, non-convex objective functions and equality and inequality requirements make the EPD issue more feasible. Recent evolutionary algorithms have improved results, especially for complex optimization problems. These methods identify global optimum solutions in non-convex and large solution domains. Recent algorithms search the solution space at random utilising numerous candidate solutions. Evolving algorithms must be able to get answers by exploring the feasible zone rather than researching the complex region. So the optimization process is speedier and consumes less calculation time while still finding the global optimum. The traditional technique fails due to premature convergence.

In this paper, consensus analysis and design are investigated for discrete-time multi-agent systems (MASs) with consecutive packet losses in communication links of directed communication topology. It is assumed that at each transmitted instant, the packet loss is stochastic and described by Bernoulli distribution, but during a given time window, the data could be successfully transmitted to the agent from each of its neighbors. First, we utilize the incidence matrix of a directed spanning tree of the communication topology to construct a linear transformation matrix so that the consensus problem is equivalently transformed into a mean square asymptotic stability problem of a reduced-order system. Second, by using Lyapunov–Krasovskii functional approach, we derive some sufficient conditions of consensus criterion in terms of linear matrix inequalities (LMIs). These conditions express the relationship among the control gain matrix, packet loss rate, and the number of consecutive packet losses. Based on the conditions, gain matrix in the consensus protocol is designed to guarantee the MAS with stochastic packet losses to achieve mean square asymptotic consensus. Moreover, for the special case of undirected communication topology, by using the matrix decomposition method, the variables of LMIs are transformed into the low-order conditions independent of the network size. Finally, some numerical examples are given to show the effectiveness of the proposed method.

Architecture, functions and decision process of the node routing system MANET have been considered. Ad Hoc networks are considered as the technical evolution from 4G towards private area networks. Mobile ad hoc net works (MANET) consist of wireless nodes that communicate with each other by cooperatively sharing of common wireless medium

Two enabling technologies for envisioned tactical network systems are mobile ad hoc network (MANET) routing and collaborative multiagent systems (MAS). Despite their respective technical value in enabling more distributed, autonomous networking, open research and engineering questions remain regarding robust interoperation, standardization, and design of these two technologies. Little work has been done to date to examine the interaction and performance of distributed agent designs within MANET environments. This paper examines the interactions and effects of running a team of belief-desire-intention (BDI) agents within a wireless network using emerging MANET protocol frameworks. The focus of the interagent communication model applied in this study is a form of MANET multicast routing and is aimed at improving group-based agent collaboration. The developed simulation testing environment is specified and results from various experiments are discussed. We present recent results examining overall MAS task performance vs. related knowledge loss induced by the underlying MANET network disruptions. We conclude by outlining several open issues and areas of further work.

?????? ???????????? ?????? ??? ???????????? ????????? ??????????????? ?????? ??????????????? ????????? ???????????? ????????? ???????????????, ???????????? ????????? ??????????????? ????????? ????????? TICN(Tactical Information Communication Network)????????? ??????????????? ???????????? ?????? ?????? ??????. TICN????????? ??? ????????? ????????? ???????????? ????????? ????????? ??????????????? ????????? ?????? ???????????????????????? ?????? ????????? ?????????, ????????? ????????? ????????? ?????? ????????? ???????????? ????????? ?????? MANET(Mobile Ad-Hoc Network) ????????? ??????????????????. ??? ??????????????? ?????? MANET ????????? ?????????????????? ?????? ???????????? ???????????? ?????? ????????? ?????????????????? ????????? ?????? ???????????? ????????????, ????????? ??????????????? ??????????????? ????????????, ?????? ???????????? ????????? ??????, ????????? ?????? ?????? ???????????? ????????? ???????????? ????????? ?????? ????????? ??????????????????, ????????? ?????? ????????? ??????????????? ?????? ?????? ??? ?????? ???????????? ?????? MANET ????????? ??????????????? ????????? ????????? ??????.