jinyujie99/wbm

wbm's Issues

Missing comparison of a closely related work

Hi,

Congratulations on your wonderful ICML work and thanks for your open-sourcing the code. When I read your paper, however, we found you missed discussion and comparison with a closely related work CIGA that achieves state-of-the-art graph size generalization performance on the same datasets and was published at NeurIPS2022. Although we understand there are distinctions in the focus of the two works, could you kindly make up the discussion of our work?

Best regards,
Yongqiang Chen

The equation (10) in the paper may be miswritten.

Hi Yujie,

Congrats to the acceptance of your new work! After reading the paper, there are two questions I hope to discuss.

• Is the equation (10) miswritten? I expect it to be like $\mathcal{O}(\log{N_S}^{1/2} * N_S^{-1/2 (D_{\chi^j} + 1)} )$ instead of $\mathcal{O}(\frac{\log{N_S}^{1/2}}{N_S^{-1/2 (D_{\chi^j} + 1)}} )$ so that the generalizability bound can be compared as described in Remark 4.12..
• Can you provide an intuitive explanation why applying Wasserstein distance based representations improves the size generalization ability? According to the ablation study between WBM and $\text{WBM}^{1/N}$, is the learnable node weights the key? I'm a bit curious about whether the node weights are fixed in EBM?

Thank you for your time. I'm looking forward to your reply!

Best,
Shurui