Advances, Systems and Applications
Author | Research Type | Problem Area | Contribution | Related Studies |
---|---|---|---|---|
Zhao et. al | Experiment | Statistical heterogeneity | A method is developed to enhance training on non-IID data by generating a restricted subset of data that is distributed globally across all edge devices [33] | [3] |
Mcmahan, et.al | Experiment | Communication cost | A realistic method for the FL is based on iterative model averaging is proposed and evaluated an exhaustive empirical evaluation [3] | |
C. T Dinh et | Experiment | Convergence analysis of FL algorithms and resource allocation | An optimization issue of resource allocation in wireless networks is addressed by proposing a FL algorithm. The goal is to capture the trade-off between the convergence time of FL and the energy consumption of UEs with heterogeneous computation and power resources [34] | |
W. Luping et. al | Experiment | Communication cost | They suggested a system called Communication-Mitigated Federated Learning (CMFL), which provides clients with feedback on the overall trend of model updates [35] | |
M. Duan et.al | Experiment | Statistical challenges in FL | They provided evidence that inaccurate FL will result from unevenly distributed training data [27] | |
S. U. Stich et.al | Experiment | Communication cost | They suggest structured updates, which would allow them to directly learn an update from a constrained space parametrized by utilizing fewer variables, thereby reducing the communication cost by two orders of magnitude [38] | |
D. C. Verma et. al | Numerical Experiment | Communication cost | When equipped with error compensation, stochastic gradient descent (SGD) with k-sparsification or compression (such as top-k or random-k) converges at the same rate as vanilla SGD, according to an evaluation of this technique that considers accumulated errors in memory |