How to reweight the input to a neural network?

1

For example, I read a set of images into a CNN. And the default weight of these images is 1.
How can I reweight these images so that they have different weights?
Can ‘DataLoader’ achieve this goal?

2

What do you really mean by the weight of an image?

3

Just like in adaboost, we need to reweight the input in each iteration. The weight is the same meaning here.

4

Well, I see two possibilities:

  1. you define a custom loss function, providing weights for each sample as you like.
  2. you repeat samples in your training set, which will result in more frequent samples having a higher weight in your loss.

As far as I know, there is no other way to do this in PyTorch, but maybe I’m wrong…

2 Likes
5

Thank you very much!