Federated learning (FL) over wireless networks suffers from significant training latency and degraded convergence due to unreliable wireless transmission, especially under blocked propagation environments. Although reconfigurable intelligent surfaces (RISs) can improve communication reliability, existing wireless FL studies rarely characterize the trade-off between learning convergence and communication delay under modulation-dependent transmission errors. In this paper, we consider a wireless FL system operating under RIS-assisted blocked-link propagation scenarios, and focus on adaptive modulation and sub-channel allocation for convergence-latency aware communication design. By characterizing the effect of symbol errors on uploaded local gradients, we derive a convergence-related upper bound that reveals the impact of symbol error rate (SER) on FL loss decay. Based on this result, we formulate a joint convergence-latency optimization problem, which is cast as a mixed-integer nonlinear programming (MINLP) problem, and solve it using a low-complexity hybrid alternating optimization framework. Extensive experiments on MNIST, CIFAR-10, and Speech Commands show that the proposed scheme consistently achieves faster convergence and higher test accuracy than existing adaptive communication schemes, especially in complex tasks and challenging wireless scenarios.
Semantic-aware edge computing has exhibited tremendous potential for reducing communication-computing-caching (3C) resource costs in vehicular networks through task-oriented semantic extraction. However, environmental dynamics and uncertainties across heterogeneous timescales pose critical challenges for 3C resource allocation in semantic-aware vehicular edge computing (VEC) networks. To this end, this paper investigates a joint semantic content caching and semantic task offloading problem in twin-timescale semantic-aware VEC scenarios, aiming to maximize the long-term utility tradeoff between task execution latency and semantic cache hit ratio. First, twin-timescale Decentralized Partially Observable Markov Decision Processes (Dec-POMDPs) are established, where semantic content caching is optimized on a large timescale, while semantic offloading and bandwidth allocation policies are learned on a small timescale. Subsequently, a novel twin-timescale 3C resource allocation solution based on multi-agent graph reinforcement learning method is proposed. Specifically, a Graphical Partial Reward Decoupling-aided Multi-Agent Proximal Policy Optimization (GPRD-MAPPO) algorithm is proposed, which incorporates graph attention networks (GAT) and credit assignment mechanism to decouple the irrelevant agents in cooperative learning by dynamically identifying inter-agent graphical dependencies. Our simulation results verify the superiority of the proposed solution in reducing task execution latency and improving semantic cache hit ratio over the benchmarks in varying numbers of vehicles and diverse task characteristics.
Yan Lin, Jinjin Shen, Yijin Zhang et al.· IEEE Transactions on Communi...· 0 citations