Skip to content

Egnn #304

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

Merged
merged 7 commits into from
Jun 15, 2022
Merged

Egnn #304

merged 7 commits into from
Jun 15, 2022

Conversation

emsal0
Copy link
Contributor

@emsal0 emsal0 commented May 23, 2022

No description provided.

@yuehhua yuehhua linked an issue May 24, 2022 that may be closed by this pull request
2 tasks
@yuehhua
Copy link
Member

yuehhua commented May 24, 2022

Would you please move codes in src/layers/conv.jl to src/layers/groups.jl?

(pr comment fixes to come later)
@yuehhua
Copy link
Member

yuehhua commented May 30, 2022

I am considering another design here. Maybe we can take two message-passing layers: one for node embedding, the other for coordinate embedding. From the paper equation (3)~(6), we can organize them into two message-passing layers:

For node embedding:

  • message function (3): $m_{ij} = \phi_e(h_i^l, h_j^l, ||x_i^l - x_j^l||^2, a_{ij})$
  • aggregate (5): $m_i = \sum_j m_{ij}$
  • update function (6): $h_i^{l+1} = \phi_h(h_i^l, m_i)$

For coordinate embedding (4):

  • message function: $y_{ij} = (x_i^l - x_j^l)\phi_x(m_{ij})$
  • aggregate: $y_i = \frac{1}{M-1} \sum_{j \ne i} y_{ij}$
  • update function: $x_i^{l+1} = x_i^l + y_i$

Thus, it form a nested message-passing layer.

@yuehhua yuehhua changed the base branch from master to new_features June 1, 2022 02:18
@yuehhua yuehhua merged commit 0378e18 into FluxML:new_features Jun 15, 2022
@yuehhua
Copy link
Member

yuehhua commented Jun 15, 2022

@emsal0 Thank you so much!

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

Successfully merging this pull request may close these issues.

Implementation of EGNN
2 participants