I have been testing federated learning with MNIST dataset and

Question

I have been testing federated learning with MNIST dataset and

using average aggregation.

when using random 8000 samples for each client i get normal improvement in local and global accuracy.

i did test to make each client to train on only one digit, i.e client_1 train on digit 5,..etc.

global accuracy is no exponential also all clients local accuracy is constant all rounds. although the compute accuracy using whole test set.

any idea why this behavior occures?

0

01.05.2024

1 ответов

28 просмотров

ⒺⓃⒹⓄⒼⒺⓃ · Accepted Answer

ⒺⓃⒹⓄⒼⒺⓃ

No idea lol but is it possible to share some code or is that overkill?

0

7 месяцев назад

Похожие чаты

I have been testing federated learning with MNIST dataset and

1 ответов

Похожие вопросы