Machine learning models can support decision-making in mobile terminals (MTs) deployments, but their training generally requires massive datasets and abundant computation resources. This is challenging in practice due to the resource constraints of many MTs. To address this issue, data-parallel distributed learning can be conducted by offloading computation tasks from MTs to the edge-layer nodes. To facilitate the establishment of trust, one can leverage trust management, say to use trust values derived from local model quality and evaluations by other nodes as access criteria. Nonetheless, security and performance considerations remain unsolved. In this paper, we propose a blockchain-assisted dynamic trust management scheme for distributed learning, which comprises nodes attributes registration, trust calculation, information saving, and block writing. The proof of stake (PoS) consensus mechanism is leveraged to enable efficient consensus among the nodes using trust values as stakes. The incentive mechanism and corresponding dynamic optimization are then proposed to further improve system performance and security. The reinforcement-learning approach is leveraged to provide the optimal strategy for nodes’ local iterations and selection. Simulations and security analysis demonstrate that our proposed scheme can achieve an optimal trade-off between efficiency and quality of distributed learning while maintaining system security.

The rise of artificial intelligence of things (AIoT) has enabled AI-powered services within wireless networks, relying on well-trained machine learning (ML) models. Distributed learning, such as federated learning (FL), allows smart devices (SDs) to collaborate on model training without sharing raw data, but privacy protection is still necessary to prevent potential information leakage from evolving attacks. Additionally, training efficiency is hampered by limited resources and selfishness of SDs. This paper considers a layered distributed learning scenario using a double-layer auction approach, where model users act as buyers, SDs act as data owners contributing their datasets, and edge layer nodes (ELNs) serve as model trainers providing computing resources. The differential privacy (DP) mechanism is utilized to add Gaussian noise to the trained models by the ELNs. Then, we formulate a joint optimization problem to optimize task assignment, data owners' sensing durations, and model trainers' local iterations and privacy budgets, aiming to maximize the utility of all participants while ensuring cost-effective and privacy-preserving distributed learning. We decompose the formulated problem into four sub-problems and design a layered algorithm to solve them and derive collaboration strategies. Simulation results validate the algorithm's performance and demonstrate the advantages of our proposed approach compared to benchmark schemes.

Federated learning (FL) has been considered as a promising paradigm for enabling distributed machine learning (ML) in wireless networks. To address the limited energy capacity of wireless devices, we propose a simultaneous wireless information and power transfer (SWIPT) aided FL, in which one FL server (FLS) co-located at a cellular base station (BS) uses SWIPT to simultaneously broadcast the global model to wireless user-devices (UDs) and provide wireless power transfer to them. The UDs then use the harvested energy to train their local models and further transmit the local models to the FLS for aggregation. To improve the spectrum efficiency, we consider that the UDs form a non-orthogonal multiple access (NOMA) group for simultaneously sending their local models over the same spectrum channel. Taking the UDs’ time-varying available energy and channel conditions into account, we propose a dynamic optimization of the UDs-scheduling, the BS's transmit-power allocation, and the UDs’ power-splitting factors for SWIPT, with the objective of minimizing the long-term energy consumption while ensuring the FL convergence. The optimization problem, however, is challenging to solve since it is a finite-horizon dynamic programming problem but with an unknown stopping time, and moreover, the action space covers both discrete and continuous variables. To address these difficulties, we first execute a series of equivalent transformations to reduce the number of decision variables and then formulate the problem as a stochastic shortest path problem, based on which we propose an actor-critic deep reinforcement learning algorithm with the proximal policy optimization to efficiently learn the policy that dynamically adjusts the UDs-scheduling for FL as well as the BS's transmit-power for SWIPT. Numerical results validate the effectiveness and performance of our proposed algorithm. The results demonstrate that our proposed algorithm can effectively reduce the long-term energy consumption in comparison with two baseline algorithms.

Federated learning (FL) has been considered as a promising paradigm for enabling distributed training/learning in many machine-learning services without revealing users’ local data. Driven by the growing interests in exploiting FL in wireless networks, this paper studies the Non-orthogonal Multiple Access (NOMA) assisted FL in which a group of end-devices (EDs) form a NOMA cluster to send their locally trained models to the cellular base station (BS) for model aggregation. In particular, we consider that the BS adopts wireless power transfer (WPT) to power the EDs (for their data transmission and local training) in each round of FL iteration, and formulate a joint optimization of the BS’s WPT for different EDs, the EDs’ NOMA-transmission for sending the local models to the BS, the BS’s broadcasting of the aggregated model to all EDs, the processing-rates of the BS and EDs, as well as the training-accuracy of the FL, with the objective of minimizing the system-wise cost accounting for the total energy consumption as well as the FL convergence latency. In spite of the strict non-convexity of the joint optimization problem, we analytically characterize the BS’s and all EDs’ optimal processing-rates, based on which we propose a layered algorithm for finding the optimal solutions for the joint optimization problem via exploiting monotonic optimization. Numerical results validate that our algorithm can achieve the optimal solution as LINGO’s global-solver (i.e., a commercial optimization package) while significantly reducing the computation-time. Moreover, the results also demonstrate that our NOMA assisted FL can reduce the system cost compared to the benchmark FL scheme with the fixed local training-accuracy by more than 70% and the conventional frequency division multiple access (FDMA) based FL by 78%.

Equipped with machine learning (ML) models, unmanned aerial vehicle (UAV) swarms can execute various applications like surveillance and target detection. However, the connections between UAVs and cloud servers cannot be guaranteed, especially when executing massive data. Thus, traditional cloud-centric approach will not be suitable, since it may cause high latency and significant bandwidth consumption. In this work, we propose a federated learning (FL) framework via non-orthogonal multiple access (NOMA) for a UAV swarm which is composed of a leader-UAV and a group of follower-UAVs. Specifically, each follower-UAV updates its local model by using its collected data, and then all follower-UAVs form a NOMA-group to send their respectively trained FL parameters (i.e., the local FL models) to the leader-UAV simultaneously. We formulate a joint optimization of the uplink NOMA-transmission durations, downlink broadcasting duration, as well as the computation-rates of the leader-UAV and all follower-UAVs, aiming at minimizing the latency in executing the FL iterations until reaching a specified accuracy. Numerical results are presented to verify the effectiveness of our proposed algorithm, and demonstrate that the proposed algorithm can outperform some baseline strategies.

Minimizing the make-span of the entire divisible load is the primary objective of multi-installment scheduling in heterogeneous parallel and distributed systems. This is a significantly difficult problem to address because we have to find the optimal number m of installments, optimal number n of processors taking part in computation, and optimal load partition A = [αij]n×mwith each element represents the load fraction assigned to each processor in different installment. Therefore, this problem involves 2+n m variables. In this paper, we first find the function expression of the optimal load partition A with respect to the number m of installments and the number n of processors participating in computation, i.e., A = f (n, m), thereby reducing the dimension of the problem down to 2. Then we propose a new heuristic method for finding the optimal numbers of installments and processors. Finally, experimental results show that the make span of the entire divisible load obtained by the proposed method is smaller than those by the existing multi-installment scheduling methods, which implies the effectiveness of the proposed method.