reproduce neural network with random seed for deep learning

I have tried rng(0) and gpurng(0) for control the random seed for deep learning. But the network is still out of control. The training returns a different network even only trained 1 epoch.

The shuffle setting use randperm function and does not effect as I understand.

Then I should promise that the result in my paper can be reproduced, and it also will effect all researchers using MATLAB for AI.

We need to know the source of the uncertainty.

Please return as soon as possible. Thanks!