In case you are interested, the code is publicly available here https://github.com/diningphil/gnn-comparison …!
-
-
Prikaži ovu nitHvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
We use the unsupervised training setting to train a GNN-based model for the graph classification task. And we show that a unsupervised model can noticeably outperform up-to-date supervised models by a large margin. Code:https://github.com/daiquocnguyen/U2GNN …
-
Thanks for sharing your work with us. We noticed you follow the same experimental procedure we criticize in our paper. Would you consider using our framework to assess your model? We think that would make a nice addition to our results!
- Još 1 odgovor
Novi razgovor -
-
-
Could you comment on this paper: Orlova, Y., Alamgir, M., von Luxburg, U.: "Graph kernel bench- mark datasets are trivial", ICML workshop '2015. They show that the datasets this work build upon are, in a sense, trivial, and graph structure does not bring any additional info 1/2
-
I have reproduced these baselines here: https://gist.github.com/xgfs/08b9123f687c638a8a6298df90c6c542 … and it seems that taking simple features with no regard to graphs' structure is enough to beat the benchmark you have presented in the paper and several GNN works.
- Još 2 druga odgovora
Novi razgovor -
Čini se da učitavanje traje već neko vrijeme.
Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.