Yijie Lin

Thanks for your interest in our work! How to conduct contrastive learning and prediction learning for more than two views is a pendent problem due to the combination problem of...

目前方法实现是每个视图一个网络进行重构的。多视图的重构损失是指用一个网络重构所有视图吗？还是说只用重建损失训练网络？

@zhangyuanyang21 训练时会通过对比损失和预测损失来学习模态间的一致性。测试时我们拼接不同视图的表示作为公共表示，然后通过kmeans验证表示学习的性能。

Thanks for your interest in our work! It is just a showcase. Our method could handle the incomplete problem of different missing rates and you can change it in configure.py.

The masked data is generated by the get_mask function and the corresponding missing rate is in configure.py. It should be pointed that the missing rate in configure.py is the real...

Thanks for your suggestions! I have changed the code to missing_rate = missing_rate / view_num now.

> Hi, I would like to ask if running a dataset with the missing rate set to 0, is it a complete view network model? Yes it is~

Exactly. You could refer to our multi-view version [DCP](https://github.com/XLearning-SCU/2022-TPAMI-DCP).

We suppose that the representation Z is continuous. Z could be understood as the distribution of discrete cluster assignment variables z.

我们的算法中图通过三角剖分构建的，详见 https://github.com/Lin-Yijie/Graph-Matching-Networks/blob/f63c575ef9c6132207a52d2285cc4583c9947456/COMMON/src/dataset/data_loader.py#L112 这里包含了三角剖分、全联通和阈值图的构建，你可以通过这个函数构建自己的图。 https://github.com/Lin-Yijie/Graph-Matching-Networks/blob/f63c575ef9c6132207a52d2285cc4583c9947456/COMMON/src/build_graphs.py#L12 要使用自己数据训练的话，可以修改train_eval.py里的dataloader。