Compression is an efficient way to relieve the tremendous communication overhead of federated learning (FL) systems. However, for the existing works, the information loss under compression will lead to unexpected model/gradient deviation for the FL training, significantly degrading the training performance, especially under the challenges of data heterogeneity and model obsolescence. To strike a delicate trade-off between model accuracy and traffic cost, we propose Caesar, a novel FL framework with a low-deviation compression approach. For the global model download, we design a greedy method to optimize the compression ratio for each device based on the staleness of the local model, ensuring a precise initial model for local training. Regarding the local gradient upload, we utilize the device's local data properties (i.e., sample volume and label distribution) to quantify its local gradient's importance, which then guides the determination of the gradient compression ratio. We have implemented Caesar, on two physical platforms with 40 smartphones and 80 NVIDIA Jetson devices. Extensive results show that Caesar, can reduce the traffic costs by about 25.54%þicksim37.88% when achieving the same target accuracy compared to the compression-based baselines, while incurring only a 0.68% degradation in final test accuracy relative to the full-precision communication.

Virtual private clouds (VPCs) play a critical role in providing secure and isolated network environments for web services. However, with the growing number and size of VPCs, efficiently delivering control messages from the control plane to the data plane has become a major concern for cloud vendors. Existing end-to-end transmission solutions (e.g., RPC) will result in substantial overhead in the control plane, while message-oriented middleware-based solutions (e.g., message queue) will lead to high data plane overhead. To address this issue, we design Meteor, a high-performance control message delivery system for large-scale clouds. Specifically, Meteor combines an RPC path with a message queue (MQ) path and employs an auto dual-path switching mechanism to minimize the message delivery latency. Additionally, we propose a VPC-based message delivery and filtering scheme for the MQ path to reduce data plane overhead. We also design a delivery robustness guarantee mechanism to ensure the reachability and consistency of control messages. Meteor has been thoroughly tested with up to 100k container instances. Evaluation results show that Meteor decreases the message delivery latency by 48.8% and reduces the overhead by about 50% in real-world scenarios, compared with state-of-the-art solutions.

Edge inference is a technology that enables real-time data processing and analysis on clients near the data source. To ensure compliance with the Service-Level Objectives (SLOs), such as a 30% latency reduction target, caching is usually adopted to reduce redundant computations in inference tasks on stream data. Due to task and data correlations, sharing cache information among clients can improve the inference performance. However, the non-independent and identically distributed (non-IID) nature of data across different clients and the long-tail distributions, where some classes have significantly more samples than others, will reduce cache hit ratios and increase latency. To address the aforementioned challenges, we propose an efficient inference framework, CoCa, which leverages a multi-client collaborative caching mechanism to accelerate edge inference. On the client side, the model is pre-set with multiple cache layers to achieve a quick inference. During inference, the model performs sequential lookups at cache layers activated by the edge server. On the server side, CoCa uses a two-dimensional global cache to periodically aggregate information from clients, mitigating the effects of non-IID data. For client cache allocation, CoCa first evaluates the importance of classes based on how frequently and recently their samples have been accessed. CoCa then selects frequently recurring classes to address long-tail distribution challenges. Finally, CoCa dynamically activates cache layers to balance lookup overhead and accuracy. Extensive experiments demonstrate that CoCa reduces inference latency by 23.0% to 45.2% on the VGG, ResNet and AST models with a slight loss of accuracy.

Kubernetes is widely adopted by enterprises to enhance service availability for applications such as web services and large-scale model training, due to its advantages in managing containerized applications. As service demands increase, a single Kubernetes cluster often becomes insufficient, leading to the trend of using multiple clusters to improve service scalability. However, achieving efficient cross-cluster communication poses significant challenges due to the need for low latency, high throughput, and strong robustness. Existing methods for cross-cluster communication either employ a centralized control plane, which becomes a communication bottleneck, or use numerous service-bound proxies, leading to increased management complexity and possibly compromised robustness in cross-cluster communication.

Federated learning (FL) provides a distributed framework for multiple clients to collaboratively train models without exposing raw data. Most FL research assumes that all clients have fully labeled data, which is impractical for many real-world applications. To this end, we focus on semi-supervised FL (SSFL), where data samples of each client are partially labeled. However, existing SSFL methods ignore two inherent characteristics of FL: limited communication resources and heterogeneous data distribution, which severely hinder convergence stability and efficiency. This paper proposes a novel SSFL mechanism, called FedAC, to address the above two challenges by alternating client-to-client (C2C) collaboration. Specifically, we group all clients using different clustering strategies at two different training stages. During each global round, FedAC first performs similarity clustering based on local data distribution, which gathers the knowledge from similar clients to generate high-quality pseudo-labels for unlabeled data. Then the clients are re-grouped using dissimilarity clustering strategy to approximate the IID setting at the cluster level, thereby alleviating the bias induced by Non-IID data. FedAC adopts a reinforcement learning algorithm to achieve a balance between labeling assistance from similar clients and unbiased optimization from dissimilar clients. Extensive evaluations demonstrate that FedAC can improve model accuracy and save up to 59.65% of communication costs compared with existing benchmarks.

Recently, federated learning (FL) has emerged as a popular technique for edge AI to mine valuable knowledge in edge computing (EC) systems. To boost the performance of AI applications, large-scale models have received increasing attention due to their excellent generalized abilities. However, training and transmitting large-scale models will incur significant computing and communication burden on the resource-constrained workers, and the exchange of entire models may violate model privacy. To relax the burden of workers and protect model privacy, split federated learning (SFL) has been released by integrating both data and model parallelism. Despite resource limitations, SFL also faces two other critical challenges in EC systems, i.e., statistical heterogeneity and system heterogeneity. In order to address these challenges, we propose a novel SFL framework, termed MergeSFL, by incorporating feature merging and batch size regulation in SFL. Concretely, feature merging aims to merge the features from workers into a mixed feature sequence, which is approximately equivalent to the features derived from IID data and is employed to promote model accuracy. While batch size regulation aims to assign diverse and suitable batch sizes for heterogeneous workers to improve training efficiency. Moreover, MergeSFL explores to jointly optimize these two strategies upon their coupled relationship to better enhance the performance of SFL. Extensive experiments are conducted on a physical platform with 80 NVIDIA Jetson edge devices, and the experimental results show that MergeSFL can improve the final model accuracy by 5.82% to 26.22%, with a speedup by about 1.39x to 4.14x, compared to the baselines.

Deep learning has brought about a revolutionary transformation in network applications, particularly in domains like e-commerce and online advertising. Distributed training (DT), as a critical means to expedite model training, has progressively emerged as a key foundational infrastructure for such applications. However, with the rapid advancement of hardware accelerators, the performance bottleneck in DT has shifted from computation to communication. In-network aggregation (INA) solutions have shown promise in alleviating the communication bottleneck. Regrettably, current INA solutions primarily focus on improving efficiency under the traditional parameter server (PS) architecture and do not fully address the communication bottleneck caused by limited PS ingress bandwidth. To bridge this gap, we propose InArt, the first work to introduce INA with routing selection in a multi-PS architecture. InArt employs a multi-PS architecture to split DT tasks among multiple PSs, and selects appropriate routing schemes to fully harness INA capabilities. To accommodate traffic dynamics, InArt adopts a two-phase approach: splitting the training model among multiple parameter servers and selecting routing paths for INA. We propose Lagrange multiplier and randomized rounding algorithms for these phases, respectively. We implement InArt and evaluate its performance through experiments on physical platforms (Tofino switches) and Mininet emulation (P4 Software Switches). Experimental results show that InArt can reduce communication time by 48%\!\sim57\!% compared with state-of-the-art solutions.

Data generated at the network edge can be processed locally by leveraging the emerging technology of Federated Learning (FL). However, non-IID local data will lead to degradation of model accuracy and the heterogeneity of edge nodes inevitably slows down model training efficiency. Moreover, to avoid the potential communication bottleneck in the parameter-server-based FL, we concentrate on the Decentralized Federated Learning (DFL) that performs distributed model training in Peer-to-Peer (P2P) manner. To address these challenges, we propose an asynchronous DFL system by incorporating neighbor selection and gradient push, termed AsyNG. Specifically, we require each edge node to push gradients only to a subset of neighbors for resource efficiency. Herein, we first give a theoretical convergence analysis of AsyNG under the complicated non-IID and heterogeneous scenario, and further design a priority-based algorithm to dynamically select neighbors for each edge node so as to achieve the trade-off between communication cost and model performance. We evaluate the performance of AsyNG through extensive experiments on a physical platform. Evaluation results show that AsyNG can reduce the communication cost by 60% and the completion time by about 30% for achieving the same test accuracy, compared to the baselines.

Federated learning (FL) has been widely adopted to train machine learning models over massive distributed data sources in edge computing. However, the existing FL frameworks usually suffer from the difficulties of resource limitation and edge heterogeneity. Herein, we design and implement FedMP, an efficient FL framework through adaptive model pruning. We theoretically analyze the impact of pruning ratio on model training performance, and propose to employ a Multi-Armed Bandit based online learning algorithm to adaptively determine different pruning ratios for heterogeneous edge nodes, even without any prior knowledge of their computation and communication capabilities. With adaptive model pruning, FedMP can not only reduce resource consumption but also achieve promising accuracy. To prevent the diverse structures of pruned models from affecting the training convergence, we further present a new parameter synchronization scheme, called Residual Recovery Synchronous Parallel (R2SP), and provide a theoretical convergence guarantee. Extensive experiments on the classical models and datasets demonstrate that FedMP is effective for different heterogeneous scenarios and data distributions, and can provide up to 4.1× speedup compared to the existing FL methods.

To approach the challenges of non-IID data and limited communication resource raised by the emerging federated learning (FL) in mobile edge computing (MEC), we propose an efficient framework, called FedMigr, which integrates a deep reinforcement learning (DRL) based model migration strategy into the pioneer FL algorithm FedAvg. According to the data distribution and resource constraints, our FedMigr will intelligently guide one client to forward its local model to another client after local updating, rather than directly sending the local models to the server for global aggregation as in FedAvg. Intuitively, migrating a local model from one client to another is equivalent to training it over more data from different clients, contributing to alleviating the influence of non-IID issue. We prove that FedMigr can help to reduce the parameter divergences between different local models and the global model from a theoretical perspective, even over local datasets with non-IID settings. Extensive experiments on three popular benchmark datasets demonstrate that FedMigr can achieve an average accuracy improvement of around 13%, and reduce bandwidth consumption for global communication by 42% on average, compared with the baselines.

Federated learning (FL) has been widely applied to collaboratively train deep learning (DL) models on massive end devices (i.e., clients). Due to the limited storage capacity and high labeling cost, there are always insufficient data stored and annotated on each client. Conversely, in cloud datacenters, there exist large-scale unlabeled data, which are easy to collect from public access (e.g., social media). Herein, upon the federated semi-supervised learning (FSSL) technology, we propose the Ada-FedSemi system, which leverages both on-device labeled data and in-cloud unlabeled data to boost the performance of DL models. Given the limited communication and massive quantity of the clients, in each training round, we decide to select partial clients to participate in FL, and their local models are aggregated by the parameter server (PS) to produce pseudo-labels for the unlabeled data, which are utilized to enhance the global model. Considering that the number of participating clients and the quality of pseudo-labels will have a significant impact on the training performance (e.g., efficiency and accuracy), we introduce a multi-armed bandit (MAB) based online algorithm to adaptively determine the participating fraction and confidence threshold during federated model training. Extensive experiments on benchmark models and datasets show that, given the same resource budget, the model trained by Ada-FedSemi achieves 3%-14.8 % higher test accuracy than that of the baseline methods. Besides, when achieving the same test accuracy, Ada-FedSemi saves up to 48% training cost, compared with the baselines.

Most user-generated data in online services are presented as set-valued data, e.g., visited website URLs, recently used Apps by a person, and etc. These data are of great value to service providers, but also bring privacy concerns if collected and analyzed directly. To tackle potential privacy threatens, local differential privacy (LDP) attracts increasing attention nowadays. However, existing approaches only provide sub-optimal error bound for set-valued data distribution estimation with LDP. Besides, it is computational expensive and communication expensive to use for high dimensional set-valued data, considering large domains in real scenarios. Thus, existing approaches are unpractical to use on resource-constrained user-side devices (e.g., smartphones and wearable devices). In this paper, we propose a utility-optimal and efficient set-valued data publication method (i.e., wheel mechanism ). On the user side, each user contributes only one numerical value to represent their privatized data. The computational complexity is O (min{ m log m , me ɛ }) and communication cost is O (log( me ɛ )) bits, while existing approaches usually depend on O ( d ) or O (log d ), where m is the number of items in the set-valued data ( m ≡ 1 for categorical data), d is the domain size (usually d ≫ m ) and ɛ is the privacy budget. On the server side, the estimator takes numerical values from users as input and derives an unbiased distribution estimation. Theoretical results show that estimation error bounds are improved from previously known [EQUATION] to the optimal rate [EQUATION]. Results on extensive experiments demonstrate that our proposed wheel mechanism is 3-100× faster than existing approaches, meanwhile has optimal statistical efficiency.

As the size of smartphone touchscreens becomes larger and larger in recent years, operability with single hand is getting worse especially for female users. We envision that user experience can be significantly improved if smartphones are able to detect the current operating hand and adjust the UI subsequently. In this paper, we propose a novel scheme that leverages user-generated touchscreen traces to recognize current operating hand accurately, with the help of a supervised classifier constructed from twelve different kinds of touchscreen trace features. As opposed to existing solutions that all require users to select the current operating hand or dominant hand manually, our scheme follows a more convenient and practical manner, and allows users to change operating hand frequently without any harm to user experience. We conduct a series of real-world experiments on Samsung Galaxy S4 smartphones, and evaluation results demonstrate that our proposed approach achieves 94.1% accuracy when deciding with a single trace only, and the false positive rate is as low as 2.6%.