Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Some ideas about parameters updating. #47

Open
zhaoyu-li opened this issue Jan 30, 2020 · 2 comments
Open

Some ideas about parameters updating. #47

zhaoyu-li opened this issue Jan 30, 2020 · 2 comments

Comments

@zhaoyu-li
Copy link

Thanks for your good implementation of MAML, however, I think that maybe using state_dict() and load_stat_dict() is much easier than modifying all the weights (in learner.py forward), can I first deepcopy the net parameters(state_dict()) and use the fast weights (also use a optimizer to update, instead of list(map(lambda p: p[1] - self.update_lr * p[0], zip(grad, self.net.parameters()))) ), then load the origin parameters back to update the meta learner? Thanks.

@im-wll
Copy link

im-wll commented Mar 31, 2020

I also think it's too complicated to redefine the initialization parameters for each layer. Is there any way to make any network (such as ResNet) put into the MAML frame without defining each layer?

@shiliang26
Copy link

I wonder if anyone has successfully implemented this, as I haven't. It appears any load operation or attempt to backprop in an alternative network would remove the computational graph.

I have been relying on redefining every layer for deeper networks so it would really help if this works.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

3 participants