Federated learning(FL) is a machine learning paradigm designed to protect data privacy and security among multiple clients. It is widely used in industries such as healthcare, finance, and insurance. However, common personalization techniques such as clustering, data augmentation, and knowledge distillation often increase privacy risks or incur high computational costs with methods like homomorphic encryption. To address these challenges, this paper proposes the Secure Personalized Federated Learning (SPFL) algorithm, combining Clustered Federated Learning (CFL) and Oblivious Transfer (OT). Using the Affinity Propagation (AP) clustering algorithm, SPFL groups clients by model similarity to create personalized global models without additional information sharing. OT has achieved efficient delivery of security models. Experimental results show SPFL enhances security, adapts to Non-Independent and Identically Distributed (Non-IID) data, improves global model accuracy by more than 6.5%, and reduces the overall running time under three alpha values by more than 19.82% compared to Paillier encryption.

Efficient task offloading is critical for Internet of Things (IoT) applications to alleviate computational load on devices and ensure real-time responsiveness. However, device heterogeneity and privacy constraints impede access to global information, limiting the accuracy of conventional offloading decisions. To address this, we propose a novel framework that integrates Case-Based Reasoning (CBR) with an Adaptive Kalman Filter (AKF) for cost estimation and task offloading. Our approach first establishes a hierarchical device structure based on transmission hops. It then leverages a historical case base to retrieve similar past instances for an initial prediction of key parameters. Subsequently, the AKF module dynamically corrects this prediction in real time, aligning the estimated cost more closely with the actual value. Upon task completion, the measured results are fed back into the case base, creating a closed-loop optimization that continuously refines the model. Experimental results demonstrate that our method significantly enhances cost estimation accuracy and offloading reliability, while effectively reducing both computational delay and energy consumption.

Opportunistic networks can support message delivery flexibly without infrastructure nor end-to-end connectivity, making them suitable for various scenarios, such as, crowd sensing, Internet of Things (IoT), vehicular ad hoc networks, etc. However, the performance of opportunistic networks, especially the delivery rate and network's operational time, is adversely affected by the uncertainty of encounter probabilities and the limited energy of IoT nodes. In this paper, we propose EQ-STAR, an optimal reachable path approach that combines encounter probabilities and energy to enhance network performance. First, the DySAT-pro model provides more accurate encounter probabilities based on the network's periodic patterns. Second, the Energy-Efficient high-Quality Markov Decision Process (EQMDP) model balances energy consumption and encounter probabilities to optimize the high-quality reachable path between the source and destination nodes. Finally, the next-hop relay node is selected based on the reachable path determined by EQMDP. Extensive experimental results show that our approach outperforms classicial and state-of-the-art baselines in both realworld and synthetic datasets.

Opportunity networks are wireless AD-hoc networks deployed in complex and harsh environments. To mitigate energy loss in opportunistic network routing, we propose a model called Nodes Sleep Scheduling based on Q-learning (NSQ) by combining the node sleep mechanism with classical opportunistic network routing algorithms when the nodes in the network have sufficient buffer space. In the NSQ model, the node sleep scheduling process is modeled as a Markov decision process, the corresponding state set, action set and reward function are defined, and the Q-learning algorithm iterates the value function. The model continuously iterates and converges through the self-learning process of the value function in the nodes, autonomously learns, and ultimately obtains an optimal sleep scheduling strategy. The experimental results show that the NSQ model can effectively reduce nodes' energy consumption and has certain self-learning abilities compared with other node sleep scheduling strategies.

Active nodes in opportunistic networks have a greater mobility range and undertake more message-forwarding tasks. This leads to issues of uneven traffic load and delayed cache space release in opportunistic networks. This paper proposes a load distribution model based on forwarding assistance (LDMFA) which integrates the node delivery prediction value and traffic idleness index to mitigate the blind selection of relay nodes and reduce message forwarding delay. To identify a selfish node, the message throughput rate of the node is selected as the evaluation index. The most appropriate relay node is selected by comparing the traffic idle index of neighbour nodes and calculating their forwarding assistance. This paper also introduces a cache optimization mechanism to address network performance degradation due to overloaded node traffic. Simulation results show that, compared with the prevailing opportunistic routing algorithms, the model improves the message delivery success ratio by 20%.

With the proliferation of mobile intelligent terminals, opportunistic networks have attracted widespread attention as a complementary technology to multi-network convergence. Different from traditional wireless networks, message delivery in opportunistic networks does not rely on a fixed infrastructure, but rather storing messages in a cache and utilizing the movement and encounters of nodes to relay messages. However, in practical application scenarios, nodes have limited storage space and energy and will easily exhibit selfishness. An increase in the number of selfish nodes will drastically degrade the performance of the network. To solve the problem of significant network performance degradation when the number of selfish nodes is high, this paper proposes an Incentive mechanism of Selfish nodes based on Energy optimization and Game Theory (ISEGT). The mechanism abstracts the process of forwarding messages by nodes into a bargaining game process, and selectively forwards messages based on nodes' remaining energy and other circumstances. The experimental results show that the ISEGT mechanism can motivate selfish nodes to actively participate in message forwarding, which improves the success rate of message delivery and the survival rate of nodes, and optimizes the overall performance of the network.

In opportunistic networks, it is a challenge to find the best relay node instead of blindly selecting from among available nodes to forward messages and effectively transmit them to their destinations. By comparing the similarity between nodes, the next hop node that is most similar to the destination node is found in time. Existing node similarity-based opportunistic networks routing algorithms only calculate similarity between nodes according to nodes' properties that can be directly obtained from the network, such as the historical meeting record between nodes, etc. However, this superficial similarity calculation is obviously inadequate to describe the inherent dynamic nature of opportunistic networks at both spatial and temporal levels, and ignores the salient movement characteristics of nodes, resulting in poor routing performance. Therefore, this paper proposes an opportunistic network routing strategy based on pre-training and fine-tuning (PTFT) model. Firstly, an autoencoder is added to the graph neural network model to encode node movement behavior. Then, a pre-training graph neural network model in large scale opportunistic network scenarios is applied to learn potential features of nodes through fine-tuning. Finally, we calculate the similarity between nodes based on the node potential feature, thereby assisting the nodes to achieve an efficient routing decision. The simulation results show that our PTFT-based algorithm not only has superior performance compared to traditional routing algorithms, but also faster learning speed than other machine learning-based routing algorithms.