1 paper · 1 filter
Chenhao Wang, Zihan Chen, Nikolaos Pappas +3
We propose a federated version of adaptive gradient methods, particularly AdaGrad and Adam, within the framework of over-the-air model training. This approach capitalizes on the in…