Is it ok to remove a loss function term by setting it's loss_weights in model.compile to zero in Keras?

Question

My total loss function has three terms:

L = λ1*L1 + λ2*L2 + λ3*L3

And all the λ are set by loss_weights{"λ1":1, "λ2":1, "λ2":1} when I run model.compile.
Now I want to remove the L1 term.
Is it ok if I change loss_weights{"λ1":0, "λ2":1, "λ2":1} in loss_weights instead of removing the output of L1 term in my model ?

Answer 1

Yes, it should be fine, it will cancel gradients from that part of the loss. This trick is commonly done in object detection losses, so we know it works.