When you applying a graph structured data to the graph convolution network, what are the benefits of using the state information that maintains the graph structure?

  • $\begingroup$ By state information, do you mean the vector representing the state of each node and, possibly, the vector representing the whole graph? $\endgroup$ – nbro Jun 27 '19 at 11:36
  • $\begingroup$ @nbro yes, the vector representing the state of such nodes information $\endgroup$ – unsmoother Jun 27 '19 at 16:44
  • $\begingroup$ Is there any specific GDL architecture you're referring to? $\endgroup$ – nbro Jun 27 '19 at 17:42
  • $\begingroup$ @nbro I recently read arxiv.org/abs/1611.08402. Although the method is compared with various GCNs methods, the method presented in this (MoNet) paper is very good. Is this because you are able to use such information better than other methods? $\endgroup$ – unsmoother Jun 27 '19 at 23:54
  • $\begingroup$ It is difficult to explain the exact reason behind the better results of MoNet w.r.t. some previous models. They use some learnable parameters, rather than handcrafting them, and they say that MoNet generalises other previous works. This might affect the performance of the model. Regarding your original question, I can try to provide an answer, but I am not sure my answer will address your issue. It's just my interpretation. $\endgroup$ – nbro Jun 28 '19 at 15:04

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.