Why I'm Lukewarm on Graph Neural Networks
singlelunch.com
Why I'm Lukewarm on Graph Neural Networks
1–10 of 50 posts
Re: Why I'm Lukewarm on Graph Neural Networks
#2Anyone have a mirror?
Re: Why I'm Lukewarm on Graph Neural Networks
#3You can use https://web.archive.org/web/20210104154304/https://www.singl... since the original website is experiencing HN bear hug.
Re: Why I'm Lukewarm on Graph Neural Networks
#4Anyone have a mirror?
Author here.
Seems like another post of mine (my take on BTC) was randomly posted and became popular at the same time I posted this one.
I'm trying to get the wordpress host's cache to handle the load.
Re: Why I'm Lukewarm on Graph Neural Networks
#5By no means I am an expert in deep learning, but lately I've been considering graph NNs snake oil of neural networks. There are no impressive results on any common tasks and that picture is supported by a recent paper claiming transformers to "contain a graph network inside".
Re: Why I'm Lukewarm on Graph Neural Networks
#6Anyone have a mirror?
Also posted on Reddit:
https://www.reddit.com/r/MachineLearning/comments/kqazpd/d_w...
Re: Why I'm Lukewarm on Graph Neural Networks
#7By no means I am an expert in deep learning, but lately I've been considering graph NNs snake oil of neural networks. There are no impressive results on any common tasks and that picture is supported by a recent paper claiming transformers to "contain a graph network inside".
Graph NNs are widely used in the chemical domain -- molecules are easily modeled as graphs, so it's a natural fit.
Re: Why I'm Lukewarm on Graph Neural Networks
#8By no means I am an expert in deep learning, but lately I've been considering graph NNs snake oil of neural networks. There are no impressive results on any common tasks and that picture is supported by a recent paper claiming transformers to "contain a graph network inside".
Graph NNs are widely used in the chemical domain -- molecules are easily modeled as graphs, so it's a natural fit.
So are graph embedding methods, which the post discusses
Re: Why I'm Lukewarm on Graph Neural Networks
#9By no means I am an expert in deep learning, but lately I've been considering graph NNs snake oil of neural networks. There are no impressive results on any common tasks and that picture is supported by a recent paper claiming transformers to "contain a graph network inside".
In my opinion there aren't many deep learning applications that are fundamentally snake-oil, it's more that people who don't quite understand the limitations and advantages of particular methods that end up being (witting or unwitting) snake-oil salespeople.
Deep graph NN's are very useful for lots of data that are inherently graph-structured (another commenter above mentioned chemistry applications, and there are lots of other examples). Whether or not for the time-being they give SOTA results on common datasets, being able to work directly with graph-structured data is quite appealing in many cases.
Re: Why I'm Lukewarm on Graph Neural Networks
#10Earlier quoted context omitted.
Graph NNs are widely used in the chemical domain -- molecules are easily modeled as graphs, so it's a natural fit.
So are graph embedding methods, which the post discusses
The post can't be accessed with a Internal Server Error. Or is it just my ISP? I often confuse myself with other representation learning methods such as graph kernels https://en.wikipedia.org/wiki/Graph_kernel
Are there any relations between them at all?