Personalized federated learning (PFL) addresses the limitations of traditional federated learning (FL) in statistically heterogeneous scenarios, where diverse client data distributions reduce model applicability. However, many PFL methods compromise client privacy or depend on external client information. This paper proposes a novel PFL method that enhances privacy and autonomy by enabling each client to share only partial model components and adaptively optimize local aggregation weights to align with its local objective. This approach minimizes reliance on other clients' information while ensuring robust personalization. Comparative experiments on three real-world datasets against seven baseline methods demonstrate that our method achieves higher or comparable accuracies, validating its effectiveness in non-IID settings.

Edge computing enables great computation ability in close proximity to the mobile devices (MDs). The task execution delay and energy consumption of the MDs will be greatly reduced by designing efficient task offloading policies. However, designing efficient task offloading policies is hard due to stochastic task arrivals. We formulate the task offloading problem as a Constrained Markov Decision Process (CMDP). We first propose a deterministic task offloading policy combining the value iteration algorithm and the sub-gradient algorithm. We then prove the existence of optimal randomized task offloading policies. Based on this, we further propose an optimal randomized task offloading policy with a closed-form policy selection probability. Simulation results demonstrate the efficiency of our algorithm in achieving low delay.

Real-time device to device (D2D) communications are important for applications of intelligent transportation, Internet of Things, etc. Most recently, researchers have focused on improving the throughput of the D2D communications underlaying cellular networks, while they have ignored the real-time requirements of packet transmissions. In this paper, we investigate the resource allocation problems for real-time D2D communications aiming to maximize the total utility of packets meeting their deadlines. First, we adopt the Markov Decision Process (MDP) to model the problem. Based on this model, we propose an optimal offline channel and slot allocation algorithm. Considering the high time complexity of the optimal offline algorithm, we then propose an online joint packet admission control, channel, and slot assignment algorithm. The online algorithm is$O(\log _2(\mu))$-competitive, where$\mu$is related to the deadlines of packets. We have proved the optimality of the online algorithm in terms of the competitive ratio among all of the online algorithms. Additionally, we have proposed a method to reduce the pessimism of the online algorithm. Simulation results show that the optimal offline algorithm achieves better performance than the online algorithm. The online algorithm outperforms the well-known real-time task scheduling algorithm EDF in terms of the total utility.