Show HN: Watch Your Step: Learning Node Embeddings via Graph Attention, NIPS'18 github.com 2 points by carlyboy 5 years ago
carlyboy 5 years ago This is a Neurips'18 workshop paper from Google AI researchers that I implemented. The reasons that I decided to implement this are as follows:0. The paper had no publicly available PyTorch implementation at the time.1. Featureless node embedding. This is the most general type of node embedding.2. Very different from current approaches -- pooling weights of implicit factorization are trainable.3. Interesting results regarding theory.4. The optimization problem itself is interesting.
This is a Neurips'18 workshop paper from Google AI researchers that I implemented. The reasons that I decided to implement this are as follows:
0. The paper had no publicly available PyTorch implementation at the time.
1. Featureless node embedding. This is the most general type of node embedding.
2. Very different from current approaches -- pooling weights of implicit factorization are trainable.
3. Interesting results regarding theory.
4. The optimization problem itself is interesting.