We have known that random seed value can affect the performance of a deep learning model.
Is there a best random seed in deep learning model or what random seed we should use when building an AI model? In this tutorial, we will discuss this topic.
How to select random seed?
In order to evaluate the affection of a random seed to AI model, one of way is to use many different random to train and evaluate the performance of AI model. Then we can find the best seed that make the AI model have best accuracy.
However, this method may take a long time to train and evaluate your model.
In paper: torch.manual seed(3407) is all you need may give us a solution.
This paper investigated the effect of random seed selection on the accuracy when using popular deep learning architectures for computer vision. From the title of this paper, we can find random seed = 3407 may make your deep learning model have a good performance.