Lightnews — Scholar-powered news

jcrwhittington.bsky.social @jcrwhittington.bsky.social · Aug 24

Come to hear about the next challenges in AI and how to solve them by an amazing set of speakers!

Thinking about Thinking @thoughtchannel.bsky.social · Aug 18

Tickets now open for the AE Global Summit on Open Problems for AI. London, 28-30th October 2025. 1,000+ leaders from AI research, policy, and industry.
@googledeepmind, @Meta , @nvidia , @ucl.ac.uk , @ox.ac.uk d
, @wayve_ai , @elevenlabs.io , and more. Link below ↓

Reposted

Ruairidh McLennan Battleday @battleday.bsky.social · Apr 24

💥 Do we now have all the components necessary to reach #AGI, or are there more discoveries to be made? Can the study of the brain help provide them?

We tackle these questions and more at our Conference on the Mathematics of Neuroscience and AI in sun-soaked Split, Croatia:

www.neuromonster.org

1 4 13

jcrwhittington.bsky.social @jcrwhittington.bsky.social · Mar 19

Great work from @bakermansjjw.bsky.social demonstrating hippocampal compositions can rapidly construct new statespaces, with replay as a potential mechanism. Indeed when a statespace changes, HPC cells are active in replay at locations they weren't active at before but are active in the future!

Jacob Bakermans @bakermansjjw.bsky.social · Mar 17

Happy to share the latest version of our work on compositional maps in hippocampus with Jo Warren, @jcrwhittington.bsky.social, @behrenstimb.bsky.social. We propose hippocampus constructs maps from cortical building blocks in replay – now with empirical support! www.nature.com/articles/s41... 1/9

Constructing future behavior in the hippocampal formation through composition and replay - Nature Neuroscience

A model of compositional state spaces in the hippocampus shows latent learning and rapid generalization. Bakermans et al. show that this model predicts the emergence of place responses in replay.

www.nature.com

2 10

Reposted

Thinking about Thinking @thoughtchannel.bsky.social · Mar 11

The program is live! discounted tickets available for students, scientists, and entrepreneurs! check tickets in the link

1 1 3

jcrwhittington.bsky.social @jcrwhittington.bsky.social · Feb 25

Great weather, a beautiful city, and a magical conference venue!

1

jcrwhittington.bsky.social @jcrwhittington.bsky.social · Feb 25

Keynote Speakers:
Prof Jay McClelland
Prof Eve Marder
Prof Wolfgang Maass
Prof Christine Constantinople
Prof @kanakarajanphd.bsky.social (@kempnerinstitute.bsky.social )
Dr Ida Momennejad @neuroai.bsky.social
Dr Kim Stachenfeld @neurokim.bsky.social
Prof Jakob Foerster

1 1 3

jcrwhittington.bsky.social @jcrwhittington.bsky.social · Feb 25

📢Abstract submissions is 16th March AOE for the Annual Conference on the Mathematics of Neuroscience and AI (May 27-30th; National Opera Theatre, Split, Croatia)

A truly outstanding line-up of keynotes, session chairs, and invited speakers. We'd love you to come along too!

www.neuromonster.org

1 2 18

jcrwhittington.bsky.social @jcrwhittington.bsky.social · Jan 16

probs relevent to the story: arxiv.org/pdf/2112.04035

1

jcrwhittington.bsky.social @jcrwhittington.bsky.social · Dec 9

Sent via email :)

1

jcrwhittington.bsky.social @jcrwhittington.bsky.social · Dec 8

Two separate models with different architectures, but their learned mechanisms - which also look different - can be seen in the same light. Don't model interactions

2

jcrwhittington.bsky.social @jcrwhittington.bsky.social · Dec 8

Thank you!

1 1

jcrwhittington.bsky.social @jcrwhittington.bsky.social · Dec 8

Many thanks to collaborators Will Dorrell @behrenstimb.bsky.social @melgaby.bsky.social @suryaganguli.bsky.social

6

jcrwhittington.bsky.social @jcrwhittington.bsky.social · Dec 8

Overall, a new unifying mechanistic theory of PFC / RNNs / SSMs, as well as a unifying understanding of cognitive maps in neural networks. (9/9)

1 4

jcrwhittington.bsky.social @jcrwhittington.bsky.social · Dec 8

Now how’s this all related to cognitive maps and hippocampus? We show that this slot algorithm is exactly the same maths as TEM just with some equations reshaped and rearranged. Frontal and temporal cognitive maps get unified! (8/9)

1 7

jcrwhittington.bsky.social @jcrwhittington.bsky.social · Dec 8

Slots don’t just have to be flat, they can be hierarchical. We show that in a hierarchical task, RNNs learn hierarchical slots! A prediction for experimentalists to test 😀. (7/9)

1 6

jcrwhittington.bsky.social @jcrwhittington.bsky.social · Dec 8

Slots also work for structured behavioural tasks too (Basu et al & El-Gaby et al). A continuous manifold of slots account for PFC neurons including ‘progress’ and ‘structured memory buffer’ neurons. These ‘behavioural’ tasks are really just working memory tasks in disguise! (6/9)

1 7

jcrwhittington.bsky.social @jcrwhittington.bsky.social · Dec 8

Armed with this theory, we can reinterpret PFC representations as slots! Not just in simple sequence repetition tasks (Xie et al), but also cue-dependent recall tasks (Panichello et al). Here the cue acts like a velocity signal to move memories between slots 😀. (5/9)

1 1 5

jcrwhittington.bsky.social @jcrwhittington.bsky.social · Dec 8

And slots are exactly what randomly initialised RNNs/SSMs actually learn! And for many many task structures and sizes (not just the one shown in the figure here) (4/9)

1 6

jcrwhittington.bsky.social @jcrwhittington.bsky.social · Dec 8

Our new slot algorithm works for any recurrent network (PFC / RNN / SSM) for any structured sequence memory task. This is a big generalisation of prior mechanisms (e.g. Botvinick or Ganguli) that could only repeat the exact sequence it was presented. (3/9)

1 5

jcrwhittington.bsky.social @jcrwhittington.bsky.social · Dec 8

What’s the new mechanism? Different memories get stored in different neural subspaces–‘slots’. Importantly, slots are connected to each other (mirroring task connectivity!) so that memories get passed between slots allowing the right memory to be readout at the right time! (2/9)

1 5

jcrwhittington.bsky.social @jcrwhittington.bsky.social · Dec 8

New paper out in #Neuron: A general theory of sequential working memory in prefrontal cortex and RNN/SSMs with their exact neural mechanism. Plus unifying this new mechanism with the alternate mechanism of hippocampal cognitive maps! (1/9)

www.cell.com/neuron/fullt...

6 40 140

Reposted

Surya Ganguli @suryaganguli.bsky.social · Dec 6

Stellar new work lead by the inimitable James Whittington in Neuron that develops a new theory unifying episodic and working memory and explains diverse hippocampal and prefrontal data: www.cell.com/neuron/fullt... w/Will Dorrell, @behrenstimb.bsky.social Mohamedy El Gaby

A tale of two algorithms: Structured slots explain prefrontal sequence memory and are unified with hippocampal cognitive maps

The algorithm of the prefrontal working memory system on sequence memory tasks is not well understood, whereas it is well understood for the hippocampal episodic memory system. This work shows a mathematical duality between working and episodic sequence memory. This leads to a computational understanding of sequence working memory—as structured activity slots. This theory algorithmically explains recurrent neural network (RNN) and prefrontal representations during sequence memory tasks.

www.cell.com

2 30 110