ardunn
commented
1 year ago Great, thanks for the submission!
When you do publish, if the algorithm differs at all from this submission, just make a new submission under an updated name! I.e., coGN-official or coGN-published or similar.
Benchmark submission
Brief description of the algorithm
We found that there is a strong interdependency between crystal preprocessing for GNNs and GNN architectures. Our model "coGN" was optimized with respect to both aspects.
The coGN model is based on the Graph Network (GN) framework and is comparably simple, as it only contains MLPs as update functions, mean or sum aggregation functions and no sophisticated message-passing scheme.
Since the corresponding paper to this model is still under review, we submit this anonymously and without any paper references. We will update this information, once the paper is publicly available.