Are You a Good Client? Client Classification in Federated Learning

Jeong, Hyejun; An, Jaeju; Jeong, Jaehoon

doi:10.1109/ictc52510.2021.9620836

Cited by 9 publications

(10 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Their reweighting strategy used iteratively reweighted least squares to integrate repeated median regression. On the other hand, there are various defense attempts to prepare for possible attacks in the form of classifying anomalies (Li et al , 2020a; Shen et al , 2016; Fang et al , 2020; Tolpegin et al , 2020; Jeong et al , 2021). Li proposed spectral anomaly detection mechanism based on models’ low-dimensional embeddings (Li et al , 2020a).…”

Section: Research Issuesmentioning

confidence: 99%

“…The aforementioned work’s objective, anomaly detection, remains the same, but several approaches leveraged clustering-based and thresholding-based approaches (Shen et al , 2016; Fang et al , 2020; Tolpegin et al , 2020; Jeong et al , 2021; Cao et al , 2020; Sun et al , 2019). Auror dealt with targeted poisoning attacks leveraging clustering and thresholding techniques.…”

Section: Research Issuesmentioning

confidence: 99%

“…In their works, however, the global server had been trained on an innocent dataset called root data set; in other words, the server did have the knowledge of benignity and malice of the client data set, which is a breach of the no-raw-data-sharing assumption. Based on Cao et al (2020) and Sun et al (2019), Jeong et al (2021) proposed anomalous and benign client classification in federated learning that leverages feature dimension reduction, dynamic clustering and cosine-similarity-based clipping to detect and classify anomalous and benign clients where the benign ones have nonIID data and IID data. Similar to the aforementioned approaches, only the benign clients’ model weights are aggregated in the global server.…”

Section: Research Issuesmentioning

confidence: 99%

See 2 more Smart Citations

Open problems in medical federated learning

Yoo

Jeong

Lee

et al. 2022

IJWIS

Self Cite

View full text Add to dashboard Cite

Purpose This study aims to summarize the critical issues in medical federated learning and applicable solutions. Also, detailed explanations of how federated learning techniques can be applied to the medical field are presented. About 80 reference studies described in the field were reviewed, and the federated learning framework currently being developed by the research team is provided. This paper will help researchers to build an actual medical federated learning environment. Design/methodology/approach Since machine learning techniques emerged, more efficient analysis was possible with a large amount of data. However, data regulations have been tightened worldwide, and the usage of centralized machine learning methods has become almost infeasible. Federated learning techniques have been introduced as a solution. Even with its powerful structural advantages, there still exist unsolved challenges in federated learning in a real medical data environment. This paper aims to summarize those by category and presents possible solutions. Findings This paper provides four critical categorized issues to be aware of when applying the federated learning technique to the actual medical data environment, then provides general guidelines for building a federated learning environment as a solution. Originality/value Existing studies have dealt with issues such as heterogeneity problems in the federated learning environment itself, but those were lacking on how these issues incur problems in actual working tasks. Therefore, this paper helps researchers understand the federated learning issues through examples of actual medical machine learning environments.

show abstract

Section: Research Issuesmentioning

confidence: 99%

Section: Research Issuesmentioning

confidence: 99%

Section: Research Issuesmentioning

confidence: 99%

See 1 more Smart Citation

Open problems in medical federated learning

Yoo

Jeong

Lee

et al. 2022

IJWIS

Self Cite

View full text Add to dashboard Cite

show abstract

“…However, previous FL research [7] often assumes that clients have fully annotated data with ground-truth labels, which can be an unrealistic assumption as labeling data is a time-consuming, expensive process that often requires the participation of domain experts. A more practical scenario is to share a limited amount of labeled data on the server while assisting clients with unlabeled data in model training [8], as proposed in recent studies.…”

Section: Introductionmentioning

confidence: 99%

“…FixMatch [9], UDA [10]), while using federated learning strategies to aggregate the learned weights. The recent method FedMatch [8] uses existing SSL methods based on pseudolabeling and enhances the consistency between predictions made across multiple models by deploying a second labeled dataset for validation on the server, but additionally increases the need for labeled data. Other approaches FedU [11], FedEMA [12], and Orchestra [13], use self-supervised strategies to correct the training results of aggregated client models by labeled data on the server.…”

Section: Introductionmentioning

confidence: 99%

FedIL: Federated Incremental Learning from Decentralized Unlabeled Data with Convergence Analysis

Yang¹,

Dong²,

Liu³

et al. 2023

Preprint

View full text Add to dashboard Cite

Most existing federated learning methods assume that clients have fully labeled data to train on, while in reality, it is hard for the clients to get task-specific labels due to users' privacy concerns, high labeling costs, or lack of expertise. This work considers the server with a small labeled dataset and intends to use unlabeled data in multiple clients for semi-supervised learning. We propose a new framework with a generalized model, Federated Incremental Learning (FedIL), to address the problem of how to utilize labeled data in the server and unlabeled data in clients separately in the scenario of Federated Learning (FL). FedIL uses the Iterative Similarity Fusion to enforce the server-client consistency on the predictions of unlabeled data and uses incremental confidence to establish a credible pseudolabel set in each client. We show that FedIL will accelerate model convergence by Cosine Similarity with normalization, proved by Banach Fixed Point Theorem. The code is available at https://anonymous.4open.science/r/fedil.

show abstract