Layer-Neighbor Sampling

[mfbalin]

Congratulations on your paper! While I was looking through it, I was disappointed to see that my efforts on GPU accelerated Layer-Neighbor Sampling were neither mentioned nor cited in your paper.

I am the author and implementer of the https://docs.dgl.ai/generated/dgl.dataloading.LaborSampler.html. My PR into DGL was merged in November 2022 and the first version of my paper was available on arXiv at the same time https://arxiv.org/abs/2210.13339v1. Since then, it has supported full GPU acceleration and UVA sampling, though multiple optimization PRs were merged into DGL until now.

Layer-Neighbor Sampling supersedes LADIES and Neighbor Sampling algorithms and is currently the state-of-the-art sampler among them. It is set to appear at NeurIPS 2023.

[Liu-rj]

Yes we have noticed your work but considering the papers was still on arXiv that time when this work is done, so we decided not to put it in our paper. However, we have discussed about the Layer-Neighbor Sampling and think actually it is a great consolidation to the motivation of gSampler's design since gSampler decouples the execution of layers but still provide holistic optimizations. We will look into your work and try to implement it using gSampler's framework and give you feedback one available. Thank your for mentioning this.:)