lorenmt / maxl

The implementation of "Self-Supervised Generalisation with Meta Auxiliary Learning" [NeurIPS 2019].

https://shikun.io/projects/meta-auxiliary-learning

lorenmt/maxl Issues

Why do you add cos_mean between grad_1 and grad_2 to the total loss?
Closed 2 years ago1
Why the \theta_2 can be updated?
Closed 2 years ago14
Is the use of second derivative in the meta objective when update theta_2 really necessary? Can we achieve the similarly performance by only using one derivative?
Closed 4 years ago4
runmulti gpus
Closed 4 years ago1
About the generated labels
Closed 4 years ago5
Why not optimize the gen_optimizer and main optimizer with a Interleaved style in one iteration just like the optimization of generator and discriminator in GAN?
Closed 4 years ago2
About train_loss2
Closed 4 years ago2
How to run this code on multiple GPUs??? Thanks!
Closed 5 years ago2