Split Learning (SL) is a promising collaborative machine learning approach, enabling resource-constrained devices to train models without sharing raw data, while reducing computational load and preserving privacy simultaneously. However, current SL algorithms face limitations in training efficiency and suffer from prolonged latency, particularly in sequential settings, where the slowest device can bottleneck the entire process due to heterogeneous resources and frequent data exchanges between clients and servers. To address these challenges, we propose the Heterogeneous Split Federated Learning (HSFL) framework, which allows resource-constrained clients to train their personalized client-side models in parallel, utilizing different cut layers. Aiming to mitigate the impact of heterogeneous environments and accelerate the training process, we formulate a latency minimization problem that optimizes computational and transmission resources jointly. Additionally, we design a resource allocation algorithm that combines the Sample Average Approximation (SAA), Genetic Algorithm (GA), Lagrangian relaxation and Branch and Bound (B&B) methods to efficiently solve this problem. Simulation results demonstrate that HSFL outperforms other frameworks in terms of both convergence rate and model accuracy on heterogeneous devices with non-iid data, while the optimization algorithm is better than other baseline methods in reducing latency.

Federated Edge Learning (FEL) emerges as a pioneering distributed machine learning paradigm for the 6 G Hyper-Connectivity, harnessing data from the IoT devices while upholding data privacy. However, current FEL algorithms struggle with non-independent and non-identically distributed (non-IID) data, leading to elevated communication costs and compromised model accuracy. To address these statistical imbalances, we introduce a clustered data sharing framework, mitigating data heterogeneity by selectively sharing partial data from cluster heads to trusted associates through sidelink-aided multicasting. The collective communication pattern is integral to FEL training, where both cluster formation and the efficiency of communication and computation impact training latency and accuracy simultaneously. To tackle the strictly coupled data sharing and resource optimization, we decompose the optimization problem into the clients clustering and effective data sharing subproblems. Specifically, a distribution-based adaptive clustering algorithm (DACA) is devised basing on three deductive cluster forming conditions, which ensures the maximum sharing yield. Meanwhile, we design a stochastic optimization based joint computed frequency and shared data volume optimization (JFVO) algorithm, determining the optimal resource allocation with an uncertain objective function. The experiments show that the proposed framework facilitates FEL on non-IID datasets with faster convergence rate and higher model accuracy in a resource-limited environment.

Federated Learning (FL) is a novel distributed machine learning approach to leverage data from Internet of Things (IoT) devices while maintaining data privacy. However, the current FL algorithms face the challenges of non-independent and identically distributed (non-IID) data, which causes high communication costs and model accuracy declines. To address the statistical imbalances in FL, we propose a clustered data sharing framework which spares the partial data from cluster heads to credible associates through device-to-device (D2D) communication. Moreover, aiming at diluting the data skew on nodes, we formulate the joint clustering and data sharing problem based on the privacy-preserving constrained graph. To tackle the serious coupling of decisions on the graph, we devise a distribution-based adaptive clustering algorithm (DACA) basing on three deductive cluster-forming conditions, which ensures the maximum yield of data sharing. The experiments show that the proposed framework facilitates FL on non-IID datasets with better convergence and model accuracy under a limited communication environment.

Online learning has been widely applied in real-time artificial intelligence (AI) applications to learn new classes from the dynamic environment. Although the deployment of AI model training over the edge can facilitate faster processing of real-time data, the learning efficiency is plagued by the limited capacity of distributed data acquisition. In fact, not all data samples are equally important, and the random data selection strategy is not beneficial to accelerate training due to redundant data processing. In this paper, we present an importance-driven data collection framework, which leverages the usefulness of important data to improve the learning efficiency over the wireless edge. Specifically, the novel model convergence metric (MCM) is constructed to evaluate the data importance dynamically for model learning. Moreover, considering the constraint of limited network resources on learning efficiency, we establish an MCM maximization problem of joint data collecting, scheduling, and feeding in an edge computing system. A two-timescale hierarchical reinforcement learning (TTHRL) algorithm is designed to decouple the original problem into two-timescale two-level subproblems, where the top-level agent is responsible for data feeding strategy in the long term and the low-level agent learns data scheduling and collecting strategy in the short term. Simulation results show that our proposed scheme can achieve better performance improvements over the baseline schemes.

As the key technology of artificial intelligence(AI), Deep Neural Networks (DNNs) have been widely used in mobile applications, such as video analytics in autonomous driving. However, due to the constrained computation capabilities on mobile devices (MDs), it is challenging to meet the critical accuracy and real-time demand of DNN tasks, which would result in a serious drop in quality of service (QoS). A popular alternative is to offload DNN tasks to edges for intelligence inference, nevertheless, this results in a heavy communication burden due to large amounts of raw data. In this paper, we propose an adaptive DNN co-Inference (ADCI) strategy which obtains the flexible computation division among devices and edge servers with elastic execution by combining the early exit and model partition policies. Establishing a balanced utility function, we jointly optimize dynamic offloading and model adoption while taking into account the multi-user and multi-server edge computing system. To tackle the high coupling among mixed variables, we propose a two-stage deep reinforcement learning (DRL) algorithm. The early-exit and model partition decisions are tracked using the Lagrange method as a soft option. Results from simulations show that the ADCI strategy performs well with timely accuracy