edge_weight and mi loss

[jiashizuo]

Thank you so much for sharing the work!but I have some doubts about the following points:

Do we not need to take the negative of the mutual information loss?Should we be minimizing the mutual information loss?

Also, in the GlobalGNN module, the calculated edge_weight does not seem to be used.

I look forward to your explanation！

非常感谢你的代码，很棒的工作！但是我对下面的几点有些疑惑，互信息损失不用取负吗，是最小化互信息损失吗，还有在globalgnn模块，所求的edge_weight，好像没有用上？期待你的解答~

[zxyzxy8888]

您好，我最近也在看这篇论文的代码，可以留一个联系方式，我们沟通一下吗？

[cnuzh]

Thank you so much for sharing the work!but I have some doubts about the following points:

Do we not need to take the negative of the mutual information loss?Should we be minimizing the mutual information loss?

Also, in the GlobalGNN module, the calculated edge_weight does not seem to be used.

I look forward to your explanation！

非常感谢你的代码，很棒的工作！但是我对下面的几点有些疑惑，互信息损失不用取负吗，是最小化互信息损失吗，还有在globalgnn模块，所求的edge_weight，好像没有用上？期待你的解答~

Thank you so much for sharing the work!but I have some doubts about the following points:

Do we not need to take the negative of the mutual information loss?Should we be minimizing the mutual information loss?

Also, in the GlobalGNN module, the calculated edge_weight does not seem to be used.

I look forward to your explanation！

非常感谢你的代码，很棒的工作！但是我对下面的几点有些疑惑，互信息损失不用取负吗，是最小化互信息损失吗，还有在globalgnn模块，所求的edge_weight，好像没有用上？期待你的解答~

感谢对我们的工作感兴趣，最小化互信息损失应该取负，global-gnn中的edge_weight已经融入到到edge_index中，具体描述见文章的III.B.1，此处的edge_weight是用在GAT模型的对比实验中。

[jiashizuo]

Thank you so much for sharing the work!but I have some doubts about the following points:
Do we not need to take the negative of the mutual information loss?Should we be minimizing the mutual information loss?
Also, in the GlobalGNN module, the calculated edge_weight does not seem to be used.
I look forward to your explanation！
非常感谢你的代码，很棒的工作！但是我对下面的几点有些疑惑，互信息损失不用取负吗，是最小化互信息损失吗，还有在globalgnn模块，所求的edge_weight，好像没有用上？期待你的解答~

Thank you so much for sharing the work!but I have some doubts about the following points:
Do we not need to take the negative of the mutual information loss?Should we be minimizing the mutual information loss?
Also, in the GlobalGNN module, the calculated edge_weight does not seem to be used.
I look forward to your explanation！
非常感谢你的代码，很棒的工作！但是我对下面的几点有些疑惑，互信息损失不用取负吗，是最小化互信息损失吗，还有在globalgnn模块，所求的edge_weight，好像没有用上？期待你的解答~

感谢对我们的工作感兴趣，最小化互信息损失应该取负，global-gnn中的edge_weight已经融入到到edge_index中，具体描述见文章的III.B.1，此处的edge_weight是用在GAT模型的对比实验中。

感谢回复！我对互信息损失还有一些疑问，文章中说希望选中的ROI权重趋于1，没有被选中的ROI权重趋于0，为什么会选用文中的损失呢？

根据代码中的实现好像是下面这种

根据代码的实现好像是X(2)经过GC层之后和X(2)互信息最大？为什么这样可以让选中的ROI权重趋于1，没有被选中的ROI权重趋于0？我实在没有搞懂，期待你的回答！