Are Hopfield networks useful?

Question

Hopfield networks have been introduced to me multiple times as a "biologically plausible" (at least relatively speaking) neural network architecture. My read on this is that they are not necessarily useful for machine learning so much as they are a curiosity because they seem to simulate the brain. Do I have the right idea here, or are Hopfield networks actually the best option for some tasks?

I'm aware that Hopfield networks have some properties which can be important - like recurrence. I'm not asking about that, just these specific networks.

Answer 1

I think you should look at the Hopfield Networks is All You Need, where they make the case that attention, as used in transformers, works similarly Hopfield networks. Also, they use it in their Modern Hopfield Networks and Attention for Immune Repertoire Classification paper for computational biology and get pretty decent results.

Are they the best at something? Well, if they really are equivalent to attention, then well, they are equal to transformers which are currently the best architectures for several language tasks like translation. This is a work in progress, but there are clearly people who are trying to make the case that they at the very least can be the best at something.