Mobile crowdsensing (MCS) enables large-scale urban data collection through ubiquitous mobile devices, but existing multi-task allocation schemes are constrained by three fundamental challenges: neglecting task similarities, leading to data redundancy; failing to address the cold-start problem on new platforms lacking historical user data; and insufficient protection of location privacy for both task publishers and sensing users. No prior work has simultaneously tackled all three. To bridge this gap, we propose a secure and efficient task-centric multi-task allocation framework. Its novelty lies in three integrated components: (i) a cosine similarity-based task clustering algorithm with a cross-validated threshold to aggregate similar tasks, enabling one data upload for multiple tasks; (ii) a cold-start-resilient user selection mechanism combining social influence propagation for user recruitment and Thompson sampling for reputation estimation of new users; and (iii) an edge-assisted Paillier homomorphic encryption protocol for bilateral location privacy preservation. Extensive experiments on the Geolife dataset demonstrate that our scheme consistently outperforms state-of-the-art baselines across various settings, achieving superior task allocation and completion rates while effectively balancing efficiency, quality, and privacy.
Remote robotic surgery requires three critical elements for patient safety i.e., instant response times, robust security measures, and precise surgical control. Modern surgical systems that combine tactile internet technology with blockchain offer promising solutions for delivering dependable, real-time healthcare services to patients remotely. Long communication delays and security vulnerabilities can compromise remote surgery. In order to tackle these problems, a secure blockchain-based architecture for tactile Internet-enabled robotic surgical environments has been proposed. Surgeons wear smart gloves capable of providing haptic feedback, enabling more intuitive interaction with remote robotic systems. This haptic sensation helps surgeons operate with greater precision and confidence. To keep the surgical system safe from unauthorized users, we use a specialized technology called a fuzzy extractor. In fuzzy extractor instead of storing passwords or security codes, the system generates unique encryption keys directly from a surgeon’s biometric data. This approach offers two major advantages the system recognizes the same surgeon every time they log in, while their raw biometric information stays completely private and protected. In addition, blockchain technology combined with smart contracts is used to securely record surgical commands and robot responses, providing transparency, immutability, and traceability of all surgical actions. Both formal and informal security analyses are conducted to validate the robustness of the proposed framework. Experimental results demonstrate the feasibility of integrating blockchain-based authentication and audit logging with low computational overhead and secure communication for telesurgery environments.
The rapid development of the Internet of Things and unmanned aerial vehicle technologies has accelerated the application of the Internet of Drones (IoD) in fields such as intelligent transportation, disaster response, and environmental monitoring. However, authentication and communication in the IoD are conducted over public channels, making them vulnerable to various attacks. Given the constraints of memory, computational capacity, and battery power, developing a lightweight and reliable authentication protocol is essential. Existing protocols still exhibit shortcomings in terms of physical security, credential updates, and resistance to privileged insider attacks. To overcome these challenges, we propose a novel authentication and key agreement protocol named ASEAKA-IoD. We integrate physically unclonable functions with fuzzy extractors to enhance the security of authentication credentials. After each authentication session, credentials are updated to further improve security. Meanwhile, we propose a dual-state anti-desynchronization mechanism and embed credential-related information in authentication requests, thereby resisting desynchronization attacks during credential updates while reducing unnecessary computations. Furthermore, to defend against privileged insider attacks from ground control stations, we design a TEE-based credential protection mechanism. Specifically, the system’s master private keys are strictly isolated within the trusted execution environment (TEE), and locally stored credentials are encrypted using keys derived from them, effectively mitigating the risks of credential leakage and identity forgery. We validate ASEAKA-IoD’s security through the real-or-random model, BAN logic, Proverif, and heuristic analysis. A comprehensive comparison between ASEAKA-IoD and related protocols indicates that the proposed protocol exhibits superior security and efficiency. Consequently, it is well-suited for the resource-constrained IoD environment.
In federated learning, heterogeneous data across clients leads to divergence between the local optimum and the global optimum. Existing work addresses this challenge by constraining local training by collecting feature anchors from clients. In this paper, we propose Federated learning with Virtual Feature Anchor (FedVFA) to coordinate local model training and achieve global consistency among local objectives. FedVFA constructs a Virtual Feature Anchor (VFA) that is independent of clients’ private data by minimizing the pairwise correlations among unit-norm anchor vectors through Gram matrix optimization on the server. The VFA serves as a common benchmark to promote alignment of feature mappings and calibration of local classifiers across local models, thereby reconciling local objectives with the global objective. Moreover, to mitigate knowledge forgetting, we generate a teacher distribution from the current local model immediately before each mini-batch update and preserve the corresponding knowledge in subsequent local training through KL divergence. Experimental results on six benchmarks under three non-IID settings validate the effectiveness of FedVFA compared with existing baselines.
Intelligent resource allocation frameworks that can concurrently optimize energy efficiency, latency, scalability, and resource utilization are necessary due to the massive number of Internet of Things applications in edge-cloud computing systems. Graph Attention-Based Reinforcement Learning for Adaptive Resource Allocation and Task Scheduling in IoT is proposed in this research. Systems (GARN-EATS). The Edge-Temporal Attention Encoder (ETAE) processes raw smart grid data by capturing multi-scale temporal patterns and edge-specific contextual information through an attention mechanism, generating enriched feature representations. The Graph-Attentive Reinforcement Network (GARN) models tasks and resources as a dynamic graph and employs multi-head graph attention with reinforcement learning to generate adaptive initial allocation decisions by considering task deadlines, resource states, and energy conditions. The optimization of these decisions is done through Pareto-based multi-objective optimization that maximizes resource usage, throughput, and fairness while minimizing task delay and energy consumption. From experimental analysis using IoT-Enabled Smart Grid Dataset, The suggested GARN-EATS model is found to have superior performance in terms of average task latency of 85ms, energy consumption of 360 Joules, resource utilization of 98%, throughput of 305 tasks per second and fairness index of 0.98 when compared to other scheduling algorithms.
