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

Concerns w.r.t. FedStar #1

Open
tdye24 opened this issue Dec 1, 2022 · 2 comments
Open

Concerns w.r.t. FedStar #1

tdye24 opened this issue Dec 1, 2022 · 2 comments

Comments

@tdye24
Copy link

tdye24 commented Dec 1, 2022

Hi,
I really appreciate the work, sharing domain-invariant structure knowledge. I have some concerns.
Although it can be seen from the ablation experiments that sharing structure encoder can indeed bring some benefits, it seems that most of the performance gains come from the decoupling mechanism[1].
In addition, the comparison with the baselines in the experiment is a bit unfair as the decoupling brings too many additional parameters compared to FedAvg.

[1] Graph Neural Networks with Learnable Struc- tural and Positional Representations

@kennencool
Copy link

Hello, I share your concerns.
I did some tentative experiments to imitate the part of Table 2 in the paper work[1], what come to the same conclusion you did. It seems that most of the performance gains come from the decoupling mechanism.
If there is a chance, we can communicate, appreciate it a lot.

[1] Graph Neural Networks with Learnable Structural and Positional Representations

@xiaxiaxiatengxi
Copy link

我比较好奇,他这个工作,为什么会有多张图呢? 按理说每一个用户的图结构不应该是一种吗?如果是多张图,比如分子信息结构……那不是和Fed+CV里的图片一样吗

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