Federated learning (FL) is a promising paradigm that can enable collaborative model training between vehicles while protecting data privacy, thereby significantly improving the performance of intelligent transportation systems (ITSs). In vehicular networks, due to mobility, resource constraints, and the concurrent execution of multiple training tasks, how to allocate limited resources effectively to achieve optimal model training of multiple tasks is an extremely challenging issue. In this paper, we propose a mobility-aware multi-task decentralized federated learning (MMFL) framework for vehicular networks. By this framework, we address task scheduling, subcarrier allocation, and leader selection, as a joint optimization problem, termed TSLP. For the case with a single FL task, we derive the convergence bound of model training. For general cases, we first model TSLP as a resource allocation game, and prove the existence of a Nash equilibrium (NE). Then, based on this proof, we reformulate the game as a decentralized partially observable Markov decision process (DEC-POMDP), and develop an algorithm based on heterogeneous-agent proximal policy optimization (HAPPO) to solve DEC-POMDP. Finally, numerical results are used to demonstrate the effectiveness of the proposed algorithm.

WiFi-based human activity recognition (HAR) holds significant application potential across various fields. To handle dynamic environments where new activities are continuously introduced, WiFi-based HAR systems must adapt by learning new concepts without forgetting previously learned ones. Furthermore, retaining knowledge from old activities by storing historical exemplar is impractical for WiFi-based HAR due to privacy concerns and limited storage capacity of edge devices. In this work, we propose ConSense, a lightweight and fast-adapted exemplar-free class incremental learning framework for WiFi-based HAR. The framework leverages the transformer architecture and involves dynamic model expansion and selective retraining to preserve previously learned knowledge while integrating new information. Specifically, during incremental sessions, small-scale trainable parameters that are trained specifically on the data of each task are added in the multi-head self-attention layer. In addition, a selective retraining strategy that dynamically adjusts the weights in multilayer perceptron based on the performance stability of neurons across tasks is used. Rather than training the entire model, the proposed strategies of dynamic model expansion and selective retraining reduce the overall computational load while balancing stability on previous tasks and plasticity on new tasks. Evaluation results on three public WiFi datasets demonstrate that ConSense not only outperforms several competitive approaches but also requires fewer parameters, highlighting its practical utility in class-incremental scenarios for HAR.

Online Class-Incremental Continual Learning (CICL) methods often rely on data replay, where a fixed-size memory buffer stores a small subset of previous data to mitigate catastrophic forgetting. However, as incremental tasks progress, the fixed buffer size reduces the number of exemplars stored per class, weakening their representativeness. We argue that preserving classification centers is more effective than maintaining decision boundaries with limited exemplars. This highlights the dual challenges of selecting high-quality exemplars for storage and addressing the imbalance between old and new classes. To tackle these issues, we propose the Representative Exploration Replay (RER) framework. Our approach evaluates exemplar representativeness using a novel metric based on the model’s classification performance, ensuring that more representative exemplars are prioritized for storage. Additionally, it mitigates class imbalance through mutual information gradient masking and knowledge distillation. Comprehensive experiments on three datasets demonstrate that RER achieves an average performance improvement of 9% over 13 state-of-the-art methods.

Federated Class-Incremental Learning (FCIL) aims to design privacy-preserving collaborative training methods to continuously learn new classes from distributed datasets. In these scenarios, federated clients face the challenge of encountering new classes while being constrained by limited memory capacity, which can lead to catastrophic forgetting in the resulting global model. Existing FCIL approaches tend to overlook the challenges posed by the heterogeneity of dataset label distribution among clients, thereby constraining the generalization capacity of the global model they learn. Some of these methods also suffer from excessive computational burdens when addressing catastrophic forgetting problems. Furthermore, certain approaches are constrained to handling only straightforward data, posing significant difficulties in managing complex datasets and tackling more intricate scenarios. In this article, we propose the Weighted Aggregation and Distillation-based FCIL (WAD-FCIL) method to address these limitations. To address data heterogeneity arising from class imbalance, we first introduce a task-aware client clustering method to identify clients with extreme class deviations before global model aggregation to eliminate potential impact on the global model. Then, we propose a multisampling weighted aggregation approach during the global FL model update that integrates knowledge from different clients and dynamically adjusts the weight of each client model to facilitate model update. To mitigate catastrophic forgetting, we propose a multimodel distillation strategy that involves selecting multiple teacher models for knowledge distillation. Extensive experiments comparingWAD-FCILwith ten state-of-the-art methods demonstrate that it significantly outperforms the baselines by 0.8%–3.2% in terms of average test accuracy on three representative benchmark datasets. The code of this work is available athttps://github.com/wufeng10010/WAD-FCIL.

In recent years, federated learning has emerged as an effective approach for the collaborative learning of decentralized data. Vertical federated learning (VFL) is a scenario of federated learning for cross-silo cooperation, where multiple parties with different features about the same set of data jointly train machine learning models without exposing their raw data. Most existing works of VFL focus on a single-job training of one machine learning model. In this paper, we propose a new framework for multi-job VFL, where multiple independent models are trained simultaneously in a cross-silo environment. We formulate the multi-job VFL scheduling problem and propose an efficient solution based on the rolling horizon method. We conduct extensive experiment to evaluate the performance of the solution. The experimental results show that our algorithm outperforms the other baseline algorithms.

The rapid development of artificial intelligence has led to a continuous expansion in the scale of the neural network models being trained, resulting in the gradual insufficiency of model training resources to support the increasingly larger models. Electric vehicles have become more intelligent thanks to their increasingly powerful computing resources. When these vehicles are not in operation, a significant amount of computing resources are left idle. This paper proposes a new distributed machine-learning system, i.e., virtual data center with parking electric vehicles (PDC), which is composed of central servers, hybrid local area networks, and electric vehicles in parking lots. The PDC is aimed at utilizing the powerful computational capabilities of electric vehicles in parking lots to provide robust computational resources for model training in favor of organizations such as universities and companies. This is accomplished by collecting jobs that need to be trained from the server, gathering information about the computation and communication resources of parking lots, and distributing the jobs that need to be trained to various electric vehicles. We further propose a reinforcement learning-based scheduling algorithm to minimize the job completion time for the jobs. The simulation results justify the feasibility of the PDC system and the efficiency of our proposed scheduling algorithm.

The integration of Artificial Intelligence (AI) and Internet of Things (IoT) devices has given rise to IoAT, promising transformative applications across various domains. Federated Learning (FL) and Split Learning (SL) are pivotal in harnessing the potential of IoAT, enabling decentralized model training while preserving data privacy. However, the heterogeneity and scalability challenges in IoAT environments necessitate advanced frameworks. In this paper, we introduce an integration of block-wise Neural Architecture Search (NAS) with a multi-partition SFL framework, called NASFLY. This approach uses only devices for actual model training and offers flexible scaling of model fragments to accommodate a wide range of device capabilities. In particular, NASFLY allows devices to utilize idle periods during the lengthy SFL forward and backward propagation phases. This is achieved by employing auxiliary model components dispatched from the server to conduct local supernet elastification using the device’s local dataset. Our method alternates between SFL for backbone network optimization and the local supernet elastification within NAS, where knowledge from the backbone network is transferred to the local supernet branches using distillation techniques. We also propose a device clustering algorithm to further improve training efficiency. Our experimental results demonstrate that this methodology significantly enhances device utilization and improves training efficiency compared with the conventional SFL.

Passively cooled base stations (PCBSs) offer low deployment cost and energy consumption for the next generation networks. By its nature, however, dealing with the thermal issue becomes crucial. For an outdoor PCBS, a major challenge is that the heat dissipation is uncertain over time. We address this online temperature scheduling problem with uncertain parameters via adjustable robust optimization (ARO) embedded into a re-optimization framework. In each optimization instance, temperature pre-scheduling is done to achieve solution robustness, looking ahead into forthcoming time slots. The solution is adaptive with respect to the gradually realized heat dissipation. Interestingly, we prove that the robust temperature pre-scheduling problem can be addressed via solving a compact linear program (LP), even though the number of possible realizations of heat dissipation is infinite. Simulation results show that our algorithm achieves robustness as well as very good average performance.

In multi-access edge computing (MEC) systems, there are multiple local cache servers caching contents to satisfy the users’ requests, instead of letting the users download via the remote cloud server. In this paper, a multi-cell content scheduling problem (MCSP) in MEC systems is considered. Taking into account jointly the freshness of the cached contents and the traffic data costs, we study how to schedule content updates along time in a multi-cell setting. Different from single-cell scenarios, a user may have multiple candidate local cache servers, and thus the caching decisions in all cells must be jointly optimized. We first prove that MCSP is NP-hard, then we formulate MCSP using integer linear programming, by which the optimal scheduling can be obtained for small-scale instances. For problem solving of large scenarios, via a mathematical reformulation, we derive a scalable optimization algorithm based on repeated column generation. Our performance evaluation shows the effectiveness of the proposed algorithm in comparison to an off-the-shelf commercial solver and a popularity-based caching.

Federated class-incremental learning (FCIL) aims to allow federated learning (FL) systems to consistently learn new tasks with classes that change dynamically, without forgetting knowledge from previous classes. In FCIL scenarios, both heterogeneity in both label and data distribution across clients and catastrophic forgetting caused by continual emergence of new classes can significantly affect the performance of a FL system. Existing FCIL methods assume only changes in class distribution over time for each single client while ignoring class-specific domain distribution. Furthermore, these methods often rely on storing old class exemplars to mitigate catastrophic forgetting, potentially raising privacy concerns and computational burdens. In this article, we propose a FCIL framework called generative federated class-incremental learning (GenFCIL) that effectively addresses the aforementioned challenges. First, we introduce a lightweight generator that promotes knowledge sharing among clients and preserves the accumulated knowledge from all clients. By collecting classes and their associated data from each client, the generator effectively tackles data heterogeneity, facilitating information transfer across clients, and mitigating catastrophic forgetting in a replay-free manner. Importantly, the lightweight nature of the generator ensures that it does not impose excessive memory and computation requirements. Second, to tackle challenges from shifts in both class distribution and class-specific domain distribution in general FCIL scenarios, which may exacerbate catastrophic forgetting, we incorporate and update multiple logit scores from clients focusing on their old and new overlapping classes to incorporate more intraclass information. Experimental results show that GenFCIL effectively alleviates the impact of catastrophic forgetting and heterogeneity.

Federated learning is a distributed machine learning framework that enables distributed edge devices to jointly train machine learning models without exchanging private data. Because federated learning can leverage edge devices’ data and computing resources, the research on federated learning has set off an upsurge. However, one of major challenges faced by federated learning is the limited computation and communication resources of edge devices, which leads to low efficiency of federated learning. In this paper, we propose a federated learning framework FedPipeline, which focuses on the local training and uploading process to improve the convergence efficiency. Different from the existing federated learning algorithms with disjoint sequential steps of training and uploading for devices, FedPipeline exploits the parallelism of training and uploading that devices upload part of the model updates in advance while they are conducting local training simultaneously. In this way, the computation and communication resources of devices can be better utilized. We further design strategies for the scheduling of early uploading and the selection of uploaded updates to maximize the benefit of FedPipeline. Experimental results show that FedPipeline can effectively accelerate the convergence speed of federated learning and preserve the accuracy of the final model.

In multi-access edge computing (MEC) systems, there are several local cache servers caching contents to satisfy the users’ requests, instead of letting the users download via the remote cloud server. In this paper, a content scheduling problem (CSP) in MEC systems is considered. Taking into account jointly the freshness of the cached contents and the traffic data costs, we study how to schedule content updates along time in a multi-cell setting. Different from single-cell scenarios, a user may have multiple candidate cache servers, and thus all cells and their caching decisions must be jointly taken. We first prove that CSP is $\mathcal{N}\mathcal{P}$-hard, then we formulate CSP using integer linear programming. For problem solving, via a mathematical reformulation, we derive a column generation algorithm embedded into a rounding scheme. Our performance evaluation demonstrates that the solutions obtained are within 0.8% from global optimality.

Caching is an effective way to address the challenges due to explosive data traffic growth and massive device connectivity in fifth-generation (5G) networks. Currently, few works on caching pay attention to the impact of the time duration for which content is stored, called retention time, on caching optimization. The research on retention time is motivated by two practical issues, i.e., flash memory damage and storage rental cost in cloud networks, together giving rise to the storage cost. How to optimize caching contents taking the storage cost into consideration is a challenging problem, especially for the scenarios with cache-enabled mobile nodes. In this paper, a retention-aware caching problem (RACP) in vehicular networks is formulated, considering the impact of the storage cost. The problem's complexity analysis is provided. For symmetric cases, an optimal dynamic programming (DP) algorithm with polynomial time complexity is derived. For general cases, a low complexity and effective retention aware multi-helper caching algorithm (RAMA) is proposed. Numerical results are used to verify the effectiveness of the algorithms.

Caching popular files at the user equipments (UEs) provides an effective way to alleviate the burden of the backhaul networks. Generally, popularity-based caching is not a system-wide optimal strategy, especially for user mobility scenarios. Motivated by this observation, we consider optimal caching with the presence of mobility. A cost-optimal caching problem (COCP) for device-to-device (D2D) networks is modeled, in which the impact of user mobility, cache size, and total number of encoded segments are all taken into account. The hardness of the problem is proved via a reduction from the satisfiability problem. Next, a lower-bounding function of the objective function is derived. By the function, an approximation of COCP (ACOCP) achieving linearization is obtained, which features two advantages. First, the ACOCP approach can use an off-the-shelf integer linear programming algorithm to obtain the global optimal solution, and it can effectively deliver solutions for small-scale and medium-scale system scenarios. Second, and more importantly, based on the ACOCP approach, one can derive a lower bound of global optimum of COCP, thus enabling performance benchmarking of any sub-optimal algorithm. To tackle large scenarios with low complexity, we first prove that the optimal caching placement of one user, giving other users' caching placements, can be derived in polynomial time. Then, based on this proof, a mobility aware multi-user algorithm is developed. Simulation results verify the effectivenesses of the two approaches by comparing them to the lower bound of global optimum and conventional caching algorithms.

Caching popular files at user equipments (UEs) provides an effective way to alleviate the burden of the backhaul networks. Generally, popularity based caching is not a system-wide optimal strategy, especially for mobility scenarios. Motivated by this observation, an optimal caching problem with respect to user mobility is investigated. To be specific, a cost-optimal caching problem (COCP) for device-to-device (D2D) networks is formulated, in which the impact of user mobility, cache size, and total number of encoded file segments are considered. Compared with the related studies, our investigation guarantees that the collected segments are non-overlapping, takes into account the cost of downloading from the network, and provides a rigorous complexity analysis. For problem solving, we first prove that the optimal caching placement of one user, giving other users' caching placements, can be derived in polynomial time. Then, based on this proof, a fast yet effective caching placement algorithm for all users is developed. Simulation results verify the effectiveness of this algorithm by comparing it to conventional caching algorithms.

In high speed rail wireless networks, handover between evolved Node Bs(eNBs) occurs frequently, and the probability of handover failure is dramatically increased, thus seriously degrading the users' experience. To tackle this issue, this paper proposes a network-assisted mobile relay (MR)- controlled fast handover scheme. Under this scheme, the serving eNB provides, in advance, the network assistance information about the available resources of the target eNB, to MR. The MR then makes the handover decision based on continual measurements. Different from the conventional handover scheme, in the proposed scheme, the handover decision is made by the MR instead of eNB. Simulation results show that this scheme can effectively improve handover performance, in terms of handover failure and communication interruption probabilities.