Self-Balancing Federated Learning With Global Imbalanced Data in Mobile Systems

Dubhe: Towards Data Unbiasedness with Homomorphic Encryption in Federated Learning Client Selection

“…In the greedy selection, the server first randomly selects a client, then continuously select clients to make the KL divergence from the data distribution of selected clients to the uniform data distribution to be minimum. Readers can refer to [8] for more details. The greedy selection method requires an overall knowledge of clients' data distribution, which is not applicable in secure FL.…”

“…FAVOR [30] uses a deep Q-learning model to select clients to maximize a reward that encourages the increase of accuracy and penalizes the use of more rounds. The framework in Astraea [8] acknowledges the access of data distribution of clients. It uses a greedy algorithm to balance data to reach the optimum in a global data imbalance setting.…”

Decentralized Deep Learning for Multi-Access Edge Computing: A Survey on Communication Efficiency and Trustworthiness

“…Moreover, a comprehensive comparison of communication efficiency between SL and FL was presented by Singh et al [33]. To study the relationship between communication efficiency and factors such as the total client number and [33] 2019 Split Learning IoT, Healthcare Thapa et al [34] 2020 Split Learning Healthcare, Image classification Khan et al [27] 2020 Stackelberg game theory Image classification Data Heterogeneity Jeong et al [35] 2018 Federated Augmentation Image classification Sener et al [36] 2018 K-Center clustering Image classification Zhao et al [37] 2018 Data-sharing strategy Image classification Wang et al [38] 2020 Reinforcement Learning Image classification Duan et al [39] 2020 Data augmentation and rescheduling Image classification Sun et al [22] 2021 Segmented Federated Learning Cybersecurity model parameter number, a trade-off between the two models was demonstrated (Fig. 4 (b)), where the hyperbola shows the regions where one model outperforms the other regarding communication efficiency, in other words, less data transmission between the client and the PS.…”

Decentralized Deep Learning for Multi-Access Edge Computing: A Survey on Communication Efficiency and Trustworthiness

“…Jeong et al [35] 2018 Federated Augmentation Image classification Sener et al [36] 2018 K-Center clustering Image classification Zhao et al [37] 2018 Data-sharing strategy Image classification Wang et al [38] 2020 Reinforcement Learning Image classification Duan et al [39] 2020 Data augmentation and rescheduling Image classification Sun et al [22] 2021 Segmented Federated Learning Cybersecurity model parameter number, a trade-off between the two models was demonstrated (Fig. 4 (b)), where the hyperbola shows the regions where one model outperforms the other regarding communication efficiency, in other words, less data transmission between the client and the PS.…”