John Hopfield
@HopfieldJohn
Followers
6K
Following
20
Media
2
Statuses
25
I work on neural networks and biology, often from the view point of my roots in physics. Professor @PrincetonNeuro.
Joined July 2020
Dense Associative Memories, aka modern Hopfield networks, have a huge memory storage capacity. But are they biologically realistic? In our new paper with @DimaKrotov we argue that they can be written in terms of biological variables.
12
109
414
I very much enjoyed reading the papers from the first iteration of this workshop in 2023. If you are working on associative memory, consider submitting your work and participating in this event.
I am super excited to announce the call for papers for the New Frontiers in Associative Memories workshop at ICLR 2025. New architectures and algorithms, memory-augmented LLMs, energy-based models, Hopfield networks, associative memory and diffusion, and many other exciting
2
9
50
This award is very hard to respond to. I have received many hundred congratulatory notes, from former students, post-docs, Princeton University juniors and seniors, funding agencies and foundations, authors, signature collectors, amateurs, elementary school neural network.
27
178
1K
RT @HildeKuehne: The CfP for our @NeurIPSConf Workshop on Associative Memory & Hopfield Networks is out! The deadline is Sep 29th!. Detail….
0
12
0
I was welcome in this group because I knew so little about conventional neuroscience that my questions threatened no one.
1
1
30
Their scientific interaction at small meetings of the Associates was intended to usefully drive the science. Alas, this meeting was more of the prima-donna behavior of a set of scientific leaders, rather than a coherent chorus, which frustrated Frank Schmitt.
1
1
17
Francis 'Frank' Schmitt already had an amazing view in 1962 of where neuroscience needed to go if you were serious about understanding higher brain function. His Neuroscience Research Program had a set of Associates spread across the relevant fields.
@TonyZador @suzanahh Yes! And some exciting things emerged out of the multidisciplinary meetings Schmidt organized on this topic - like @HopfieldJohn ’s inspiration to create Hopfield networks.
1
7
35
The mathematical idea that makes this possible is the Lagrangian for the dynamical system. This system can be turned into a multi-layer network with simultaneous forward and feedback flow of information.
1
5
26
Everybody knows about Lyapunov functions in simple networks with local non-linearities. What this paper shows is that with symmetric feedback a Lyapunov function can be constructed for arbitrary non-linear activation functions involving groups of neurons.
One of the great features of deep learning is that we can easily stack multiple layers (e.g. dense, conv, attention) with arbitrary activation functions to build a useful feedforward network. Wouldn’t it be cool if we could do the same for Modern Hopfield Networks with feedback?
2
28
124
Very much looking forward to hearing this dialog Wednesday.
1
7
43
A perceptive commentary on the Biden election and the US political divide from the English experience of Thatcherism and Brexit. On with the celebration via @JARWillis.
0
1
11
A feeling of community is so important!.
Hello world! Welcome to the brand new twitter home of APS-DBIO! Connecting members (including potential future members) of APS-DBIO. Please help us grow, follow + retweet! @MARGARETSCHEUNG @squishycell1.@DrJennyRoss @ApsDsoft @wbialek @oritpeleg @phybiofunc @cplcillinois.
0
2
15
RT @DimaKrotov: New microscopic theory of Dense Associative Memory aka modern Hopfield network can be reduced to the model proposed in “Hop….
0
58
0
RT @DimaKrotov: Several people asked me today about Dense Associative Memories and Hopfield networks. In this talk from 2018 I explain some….
0
15
0
RT @DimaKrotov: New blog for @icmlconf paper. Fruit fly olfactory network inspires unsupervised similarity search. Data driven locality sen….
0
22
0
New @icmlconf paper on biologically inspired unsupervised similarity search with @wrong_whp @DimaKrotov @LeopoldGrinberg #icml2020
0
34
88