is the code corresponding to the algorithm in paper ?

[aliutkus]

Hi, thanks for providing code for your paper

I was wondering, when you do the actual update there:

MultiObjectiveOptimization/multi_task/train_multi_task.py

Line 184 in d45eb26

optimizer.step()

It looks to me that you are not exactly performing the same thing as what's stated in the paper
https://proceedings.neurips.cc/paper/2018/file/432aca3a1e345e339f35a30c8f65edce-Paper.pdf

On line 2, the "heads" parameters are updated through theta_t = theta_t - eta * grad, but here you do theta_t = theta_t * scale_t * eta * grad
is that correct ?

best

antoine

[aliutkus]

ups, I realize it's the same issue as
#12

closing this