A Robust Client Selection Mechanism for Federated Learning Environments

Federated learning (FL) represents a transformative shift in machine learning, moving from conventional centralized approaches to a distributed framework that emphasizes data privacy and security. The FL server transmits an initial model to clients, which they train locally on their private data. After training, the server aggregates model updates from each client to update the global model. Selecting the best clients in FL is critical to improving the convergence speed and accuracy of the final model, which requires careful client selection approaches. The client selection phase of FL faces numerous challenges that impact overall training performance, including statistical heterogeneity and system heterogeneity resulting from the diversity of client data and resources. Communication costs present another challenge, especially in networks with limited client communication resources. Additionally, selecting trustworthy clients represents another challenge, as selecting malicious clients creates a significant risk within the FL training process. Moreover, the fairness challenge entails providing fair opportunities for all clients to participate in training. To address these challenges, we offer solutions that utilize effective techniques, approaches, and client selection methods. This survey presents a taxonomy of modern client selection methods in FL, highlighting the improvements in FL performance and effectiveness achieved through these methods, including greedy selection, reinforcement learning-based selection, multi-armed bandit-based selection, clustering-based selection, and reputation & security-based selection. Subsequently, it offers a general comparison between these methods in terms of their core ideas, advantages, limitations, and use cases. Finally, future and potential trends in client selection and improving performance in FL are identified.

Federated learning (FL) is known to be susceptible to model poisoning attacks in which malicious clients hamper the accuracy of the global model by sending manipulated model updates to the central server during the FL training process. Existing defenses mainly focus on Byzantine-robust FL aggregations, and largely ignore the impact of the underlying deep neural network (DNN) that is used to FL training. Inspired by recent findings on critical learning periods (CLP) in DNNs, where small gradient errors have irrecoverable impact on the final model accuracy, we propose a new defense, called a CLP-aware defense against poisoning of FL (DeFL). The key idea of DeFL is to measure fine-grained differences between DNN model updates via an easy-to-compute federated gradient norm vector (FGNV) metric. Using FGNV, DeFL simultaneously detects malicious clients and identifies CLP, which in turn is leveraged to guide the adaptive removal of detected malicious clients from aggregation. As a result, DeFL not only mitigates model poisoning attacks on the global model but also is robust to detection errors. Our extensive experiments on three benchmark datasets demonstrate that DeFL produces significant performance gain over conventional defenses against state-of-the-art model poisoning attacks.

Trust driven On-Demand scheme for client deployment in Federated Learning

Federated learning (FL) has become a promising technology to provide edge Artificial Intelligence (AI) due to its advantages in privacy protection and reduced communication costs. However, FL is still confronted with privacy leakage issues because the sharing local model may expose the training data information. Existing works typically utilize secure aggregation techniques to eliminate privacy leakage, where local model parameters in FL are obfuscated before they are sent to the aggregator. Nevertheless, secure aggregation makes poisoning attacks more convenient given that existing anomaly detection methods mostly require access to plaintext local models. A Poisoning-Defense Secure Aggregation in FL (PDSA-FL) is proposed to enhance the privacy protection of honest clients and defend against poisoning attacks from malicious clients. Specifically, a Secure Aggregation scheme based on Random Parameters Decomposition (SARPD) is designed to protect client privacy during the FL aggregation process and eliminates the impact of dropped clients on the aggregation results. Secondly, a Poisoning Detection method based on Similarity Grouping (PDSG) is proposed to mitigate the impact of poisoning attacks on the global model of FL without leaking client model parameters. The security analysis discusses the effectiveness of the proposed PDSA-FL in terms of privacy protection. Extensive simulation results show that PDSA-FL can effectively defend against poisoning attacks, significantly improve the convergence performance of global models, and reduce the computation time of clients.

Federated learning (FL) enables privacy-preserving distributed machine learning by training models on edge client devices using their local data without revealing their raw data. In edge environments, various applications require different neural network models, making it crucial to perform joint training of multiple models on edge devices, known as multi-task FL. While existing multi-task FL approaches enhance resource utilization on edge devices through adaptive resource configuration or client selection, optimizing either of these aspects alone may lead to suboptimality. Therefore, in this paper, we explore a joint client selection and resource configuration method called JCSRC for multi-task FL, aiming to maximize energy efficiency in environments with limited computation and communication resources and heterogeneous client devices. Firstly, we formalize this problem as a mixed-integer nonlinear programming problem considering all these characteristics and prove its NP-hardness. To address this problem, we first design a multi-agent reinforcement learning (MARL)-based client selection method that selects appropriate clients for each task to train their models. The MARL method makes client selection decisions based on the clients’ data quality, energy efficiency, communication, and computation capacity to ensure fast convergence and energy efficiency. Then, we design a particle swarm optimization (PSO)-based resource configuration scheme that configures appropriate computation and bandwidth resources for each task on each client. The PSO scheme makes resource configuration decisions based on theoretically derived optimal CPU frequency and bandwidth to achieve high energy efficiency. Finally, we carry out extensive simulations and testbed-based experiments to validate our proposed JCSRC. The results demonstrate that, in comparison to state-of-the-art solutions, JCSRC can save energy consumption by up to 59% to achieve the target accuracy.

Utilizing the federated learning (FL) technique, data owners can collaboratively train artificial intelligence models, retaining all training data on their premises to minimize the potential for personal data breaches. However, self-interested users (e.g., free riders) bring new challenges that hinder the development of FL techniques. To this end, we propose a two-stage client selection scheme comprising a multiarmed bandit (MAB)-based candidate client selection method and an auction-based training client selection method. Specifically, our client selection scheme initially formulates the FL system into an MAB system, where clients are the arms and the server is the player. Then, we quantify the similarity between a local model and the server side, which is the designed metric for model aggregation and reward computation updating based on the fuzzy mathematical strategy. Next, based on the Thompson Sampling strategy, the server can intelligently determine the reward of each client, and clients with more significant rewards have the chance for local model training. With an auction method, the server can determine the training clients to reduce the training cost while maximizing each client’s revenue. Extensive experiments on real-world data sets demonstrate that the proposed scheme outperforms representative FL schemes (i.e., FedAvg, FedProx, FedMax, and MFL) regarding the model’s convergence rate and cost in FL systems with free riders.

Federated learning (FL) unleashes the full potential of training a global statistical model collaboratively from edge clients. In wireless FL, for the scarcity of spectrum, only a fraction of clients are capable to participate in the FL training in each round. On the other hand, the performance of FL suffers from the label noise, which naturally exists in the dataset of each client. To alleviate the label noise issue, prior works proposed several client selection algorithms, where the privacy of raw data might be compromised when extra information exchange was introduced. To avoid extra information exchange in FL, we propose to leverage an algorithm that measures the noise ratio of each client based on a clean validation dataset. We then propose an online client selection framework, supported by Copeland score and multi-arm bandits, which can select the client set with low noise ratios efficiently. The proposed algorithm is performed in the server side, thereby it can be implemented in the basic FL framework seamlessly. Simulation results demonstrate that our proposed algorithm spends less training time while guaranteeing the required accuracy compared to other baseline algorithms.

Non-intrusive load monitoring (NILM), as a key technology in smart-grid advanced metering infrastructure, aims to disaggregate mains power from smart meters into individual load-level power consumption. Traditional NILM methods require centralizing sensitive measurement data from users, which poses significant privacy risks. Federated learning (FL) enables collaborative training without centralized measurement data, effectively preserving privacy. However, FL-based NILM systems face serious threats from attacks such as model inversion and parameter poisoning, and rely heavily on the availability of a central server, whose failure may compromise measurement robustness. This paper proposes a hybrid FL framework that dynamically switches between centralized FL (CFL) and decentralized FL (DFL) modes, enhancing measurement privacy and system robustness simultaneously. In CFL mode, layer-sensitive pruning and robust parameter aggregation methods are developed to defend against model inversion and parameter poisoning attacks; even with 30% malicious clients, the proposed defense limits the increases in key error metrics to under 15.4%. In DFL mode, a graph attention network (GAT)-based dynamic topology adapts to mitigate topology poisoning attacks, achieving an approximately 17.2% reduction in MAE after an attack and rapidly restoring model performance. Extensive evaluations using public datasets demonstrate that the proposed framework significantly enhances the robustness of smart-grid measurements and effectively safeguards measurement privacy.

Federated Learning (FL) is an emerging distributed learning framework for collaboratively training global models while protecting data privacy. However, FL faces serious threats from malicious attacks such as model poisoning and data poisoning. To address these challenges, this paper proposes FedMRA (MRA: Multimodal feature analysis and Reputation Aggregation). FedMRA establishes an anomaly detection index system through multimodal feature extraction, combines HDBSCAN clustering, historical trajectory sparsification analysis and dynamic reputation assessment mechanism, and adopts an aggregation strategy based on reputation weight fusion to defend against malicious client damage. Experiments show that FedMRA performs well against multiple types of attacks (label flipping attack, gaussian attack, and random gradient attack), especially when up to 30% malicious clients are involved, and still significantly improves the robustness and accuracy of the global model. FedMRA provides an efficient and robust defense scheme for FL, and it performs well in both no-attack and attack scenarios.

Federated Learning (FL) has gained significant attention as a distributed machine learning paradigm, leveraging edge clients for collaborative training while preserving data privacy. However, statistical heterogeneity, arising from diverse client data distributions, poses a significant challenge to FL performance. In this work, we propose a novel heterogeneity-aware client selection scheme based on prototype margin to mitigate these challenges. By utilizing prototype learning, we capture effective data representations, allowing client selection to better account for both data quality and distribution. The proposed scheme is designed for practical wireless communication networks, where clients are clustered and we select participants based on joint prototype margin and wireless channel. To further optimize the training process, we jointly optimize client selection and bandwidth allocation, accounting for dynamic network conditions and heterogeneous client capabilities. Experimental results demonstrate that our approach significantly improves model accuracy and reduces communication overhead.

The challenges of distributed computing and privacy preservation in the Internet of Things (IoT) can be effectively addressed through the federated learning (FL) paradigm. A pivotal aspect of this approach is client selection, which plays a crucial role in identifying participating clients while balancing model accuracy and communication efficiency. However, existing methods often have limitations, such as an inability to quickly and accurately assess data quality on the client side and a failure to account for the interdependence among clients by treating each as an independent contributor. To address these issues, we introduce GPFL, a novel framework for client selection in FL. We propose an indicator that rapidly and accurately measures the contribution of heterogeneous client data to FL. Additionally, we design an exploration–exploitation mechanism that selects the optimal combination of clients while considering their interdependencies. Both theoretical and experimental analyzes demonstrate the effectiveness of GPFL in improving FL efficiency and accuracy while preserving data privacy. Experiments on federated extended MNIST (FEMNIST) and CIFAR-10 show that GPFL outperforms baseline methods in Non-IID scenarios, achieving an average improvement of over 9% in FEMNIST accuracy compared to SOTA preselection methods. Our code is available at <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://figshare.com/articles/dataset/PBFL/29264663</uri>.

Various computer vision techniques based on deep neural networks have been proposed to detect objects accurately and fast. However, due to the privacy, security and communication bandwidth restrictions of diverse participating parties, it is sometimes prohibitive to train such models on a centralized machine. Federated Learning (FL) provides a promising solution to learn a model from decentralized data. Despite the advances in FL, the diversity of client regions in which they operate and the Non-IID nature of the crowdsourced datasets reduces the accuracy of object detection models significantly. In this paper, we introduce a novel FL object detection system to efficiently train models with heterogeneous client datasets. We propose lightweight client selection methods to learn object detection models faster. Our client selection methods based on the object data distribution at clients achieves up to 74% reduction in required federated rounds compared to conventional approaches. We further extend this method by leveraging the metadata of the training images (e.g., location, direction, depth), to select clients which maximize the coverage of diverse geographical regions. We report on extensive experiments with real datasets.

Secure aggregation is a critical component in federated learning (FL), which enables the server to learn the aggregate model of the users without observing their local models. Conventionally, secure aggregation algorithms focus only on ensuring the privacy of individual users in a single training round. We contend that such designs can lead to significant privacy leakages over multiple training rounds, due to partial user selection/participation at each round of FL. In fact, we show that the conventional random user selection strategies in FL lead to leaking users' individual models within number of rounds that is linear in the number of users. To address this challenge, we introduce a secure aggregation framework, Multi-RoundSecAgg, with multi-round privacy guarantees. In particular, we introduce a new metric to quantify the privacy guarantees of FL over multiple training rounds, and develop a structured user selection strategy that guarantees the long-term privacy of each user (over any number of training rounds). Our framework also carefully accounts for the fairness and the average number of participating users at each round. Our experiments on MNIST, CIFAR-10 and CIFAR-100 datasets in the IID and the non-IID settings demonstrate the performance improvement over the baselines, both in terms of privacy protection and test accuracy.

Federated learning effectively addresses the issues of data privacy and communication overhead in traditional deep learning through distributed local training. However, its open architecture is seriously threatened by backdoor attacks, where malicious clients can implant triggers to control the global model. To address these issues, this paper proposes a novel three-stage defense mechanism based on local clients. First, through text readability analysis, each client’s local data is independently evaluated to construct a global scoring distribution model, and a dynamic threshold is used to precisely locate and remove suspicious samples with low readability. Second, frequency analysis and perturbation are performed on the remaining data to identify and disrupt triggers based on specific words while preserving the basic semantics of the text. Third, n-gram distribution analysis is employed to detect and remove samples containing abnormally high-frequency word sequences, which may correspond to complex backdoor attack patterns. Experimental results show that this method can effectively defend against various backdoor attacks with minimal impact on model accuracy, providing a new solution for the security of federated learning.

Owing to its practical configuration to edge computing and privacy preservation capabilities, federated learning (FL) has been increasingly appealing in Internet of Things (IoT) networks. However, due to the inherent openness of IoT network architectures, FL clients are susceptible to various attacks, resulting in unreliable local model updates. To address this challenge, federated unlearning (FU) emerges as a viable solution, which can erase such unreliable updates from the FL model using the unlearning operation while preserving model accuracy. Existing FU studies have significant potential, but they are not directly applicable to IoT networks because of their high computational costs and limited capacity to defend against prevalent dynamic attacks in mobile network environments. In this work, we propose FedRemover, a novel FU method specifically tailored for deployment in IoT networks. The key insight behind FedRemover is that model updates will exhibit inconsistency when exposed to attacks. Therefore, we devise a real-time malicious client detection scheme by examining the performance consistency of model updates. Upon detecting malicious clients, FedRemover promptly executes the unlearning operation, achieving an unlearned global model within a minimal number of rounds. This makes FedRemover highly efficient and robust against dynamic attacks, enabling it well-suited for practical deployment in IoT networks. Experiments on three standard datasets demonstrate the efficiency and robustness of FedRemover, with an obvious speed-up of 10× and comparable robustness guarantees compared with benchmark algorithms.