Hopfield Networks Is All You Need (2020)(arxiv.org)
arxiv.org
Hopfield Networks Is All You Need (2020)
https://arxiv.org/abs/2008.02217
https://github.com/ml-jku/hopfield-layers
7 comments
So far I’ve needed hundreds of “all you needs” beginning to think papers with that title are not all I need… Let’s try and get a bit more creative than stealing the same name scheme used when Transformers were introduced.
Tech authors seem to absolutely love hackneyed titles.
We're lucky it isn't "Falsehoods programmers believe about the unreasonable effectiveness of hopfield networks considered harmful, for fun and profit."
We're lucky it isn't "Falsehoods programmers believe about the unreasonable effectiveness of hopfield networks considered harmful, for fun and profit."
I remember when this came out. It seemed cute, but appeared to have very little impact long-term unless I'm mistaken. Have there been any major developments from this?
Yann LeCun (Meta AI head) is still all-in on energy based models, which are basically Hopfield-like. Not sure if there are any major developments in that direction, but the hype/optimism is still there. https://iopscience.iop.org/article/10.1088/1742-5468/ad292b/...
Related:
Hopfield Networks Is All You Need - https://news.ycombinator.com/item?id=26990000 - April 2021 (40 comments)
Hopfield Networks Is All You Need - https://news.ycombinator.com/item?id=26990000 - April 2021 (40 comments)
Gemini is good enough ?