Auxiliary loss

Open a18700 opened this issue 5 years ago • 1 comments

Hello, could you elaborate the rationale behind the adoption of link prediction loss: nearby nodes to be pooled together?

It's a bit confusing.

Mar 27 '20 08:03 a18700

My guess is that link prediction loss encourages nearby nodes to have links between them by encouraging their embeddings to be similar. On the other hand, the pooling layers use softmax for assigning nodes to clusters => similar embeddings are assigned to similar clusters. Then, link prediciton loss encourages nearby nodes to be pooled together.

Jan 20 '21 04:01 phucdoitoan