Lightnews — Scholar-powered news

Reposted by Jesse Geerts

Jesse Geerts @jessegeerts.bsky.social · 20h

Remember to submit an abstract to UCL NeuroAI 2025 by Wednesday! 🧠🤖 This will be a great opportunity to present your work to the local community, and we have a fantastic line-up of keynote speakers:

@ucl-neuroai.bsky.social @sainsburywellcome.bsky.social

UCL NeuroAI @ucl-neuroai.bsky.social · 6d

Submission deadline extended!

Submit your poster or lightning talk abstract by Wednesday 15th October 2025, 12:00pm

UCL NeuroAI @ucl-neuroai.bsky.social · 17d

📢 We are very happy to announce the 2025 @ucl.ac.uk
#NeuroAI Annual Conference!

Join us on Wed 5 November 2025 at @uclchildhealth.bsky.social

Details and registration: www.ucl.ac.uk/event-ticket...
Poster and lightning talks submissions are now open, up to 8th Oct.

@ucl-neurodomain.bsky.social

1 5 7

Jesse Geerts @jessegeerts.bsky.social · 20h

Abstracts are only 200 words, so if you're working on a Cosyne abstract simply submit a shortened version of your summary at the page below:

www.ucl.ac.uk/event-ticket...

UCL Event Ticketing

www.ucl.ac.uk

Jesse Geerts @jessegeerts.bsky.social · 20h

- @leenacvankadara.bsky.social (UCL)
- Netta Cohen (Uni of Leeds)
- @docqhuys.bsky.social (UCL)
- @kenneth-harris.bsky.social (UCL)
- @laklab.bsky.social (Oxford)
- @nathanieldaw.bsky.social (Princeton)

1

Jesse Geerts @jessegeerts.bsky.social · 20h

Remember to submit an abstract to UCL NeuroAI 2025 by Wednesday! 🧠🤖 This will be a great opportunity to present your work to the local community, and we have a fantastic line-up of keynote speakers:

@ucl-neuroai.bsky.social @sainsburywellcome.bsky.social

UCL NeuroAI @ucl-neuroai.bsky.social · 6d

Submission deadline extended!

Submit your poster or lightning talk abstract by Wednesday 15th October 2025, 12:00pm

UCL NeuroAI @ucl-neuroai.bsky.social · 17d

📢 We are very happy to announce the 2025 @ucl.ac.uk
#NeuroAI Annual Conference!

Join us on Wed 5 November 2025 at @uclchildhealth.bsky.social

Details and registration: www.ucl.ac.uk/event-ticket...
Poster and lightning talks submissions are now open, up to 8th Oct.

@ucl-neurodomain.bsky.social

1 5 7

Reposted by Jesse Geerts

Weijia Z @weijiazh.bsky.social · 6d

🚀 New Preprint from our team: comparing place cells across species!
Disentangling methods from biology provides a roadmap for cross-species insights into spatial coding 🌍
👉 www.biorxiv.org/content/10.1...

3 6 10

Reposted by Jesse Geerts

UCL NeuroAI @ucl-neuroai.bsky.social · 6d

Submission deadline extended!

Submit your poster or lightning talk abstract by Wednesday 15th October 2025, 12:00pm

UCL NeuroAI @ucl-neuroai.bsky.social · 17d

📢 We are very happy to announce the 2025 @ucl.ac.uk
#NeuroAI Annual Conference!

Join us on Wed 5 November 2025 at @uclchildhealth.bsky.social

Details and registration: www.ucl.ac.uk/event-ticket...
Poster and lightning talks submissions are now open, up to 8th Oct.

@ucl-neurodomain.bsky.social

UCL Event Ticketing

www.ucl.ac.uk

5 8

Reposted by Jesse Geerts

UCL NeuroAI @ucl-neuroai.bsky.social · 17d

📢 We are very happy to announce the 2025 @ucl.ac.uk
#NeuroAI Annual Conference!

Join us on Wed 5 November 2025 at @uclchildhealth.bsky.social

Details and registration: www.ucl.ac.uk/event-ticket...
Poster and lightning talks submissions are now open, up to 8th Oct.

@ucl-neurodomain.bsky.social

UCL Event Ticketing

www.ucl.ac.uk

2 12

Reposted by Jesse Geerts

Kris Jensen @kristorpjensen.bsky.social · 19d

I’m super excited to finally put my recent work with @behrenstimb.bsky.social on bioRxiv, where we develop a new mechanistic theory of how PFC structures adaptive behaviour using attractor dynamics in space and time!

www.biorxiv.org/content/10.1...

9 85 210

Reposted by Jesse Geerts

Andrej Bicanski @andrejbicanski.bsky.social · Sep 8

Dear colleagues, we have an open PhD position in computational neuroscience - spatial memory models and intracranial recordings - fully funded. Re-advertised, candidates are ideally available in the near future. A collab. with 4 labs. Apply here: tinyurl.com/yc84ctap

PhD position or a Doctoral Candidate in computational neuroscience. (m/f/d) | Karriereportal Max-Planck-Institut für Kognitions- und Neurowissenschaften

tinyurl.com

2 8 15

Reposted by Jesse Geerts

Neuroverse @neuroversepod.bsky.social · Aug 28

Check out our latest episode on habit formation with DrFrancesca Greenstreet ✅📝 We talk about how habits are made how they may not require reward-based learning … 🎧 open.spotify.com/episode/1gZI...

1 1 2

Reposted by Jesse Geerts

Daniel Wurgaft @danielwurgaft.bsky.social · Jun 28

🚨New paper! We know models learn distinct in-context learning strategies, but *why*? Why generalize instead of memorize to lower loss? And why is generalization transient?

Our work explains this & *predicts Transformer behavior throughout training* without its weights! 🧵

1/

2 7 47

Reposted by Jesse Geerts

Ching Fang @chingfang.bsky.social · Jun 26

Humans and animals can rapidly learn in new environments. What computations support this? We study the mechanisms of in-context reinforcement learning in transformers, and propose how episodic memory can support rapid learning. Work w/ @kanakarajanphd.bsky.social : arxiv.org/abs/2506.19686

From memories to maps: Mechanisms of in context reinforcement learning in transformers

Humans and animals show remarkable learning efficiency, adapting to new environments with minimal experience. This capability is not well captured by standard reinforcement learning algorithms that re...

arxiv.org

3 24 73

Jesse Geerts @jessegeerts.bsky.social · Jun 17

Thank you! Yes, I think that’s a fair summary. Another way of looking at it is that pre training on a match and copy task gives it a hint in the “wrong” direction. Our takeaway is that what the transformer learns to implement in-context depends on the pretraining task

2

Reposted by Jesse Geerts

Kim Stachenfeld, PhD @neurokim.bsky.social · Jun 16

New work on relational reasoning in transformers!

TLDR: Inductive biases of In-Weight and In-Context Learning in transformers are really different for relational reasoning, and pretraining can make a big difference for in-context.

Check out @jessegeerts.bsky.social's thread for more!

1 6 27

Reposted by Jesse Geerts

Sainsbury Wellcome Centre @sainsburywellcome.bsky.social · Jun 11

‘Some dopamine neurons signal default behaviors to reinforce habits’

SWC research features in @thetransmitter.bsky.social ⬇️

www.thetransmitter.org/neurobiology...

Dopamine neurons signal default behavior to reinforce habits

Movement-sensing neurons in the striatum influence a mouse’s choice of action by favoring routine behaviors.

www.thetransmitter.org

1 3

Jesse Geerts @jessegeerts.bsky.social · Jun 6

Check out the Psych Review paper here: psycnet.apa.org/fulltext/202...

APA PsycNet

psycnet.apa.org

3

Jesse Geerts @jessegeerts.bsky.social · Jun 6

This is a nice paper which applies and refines some of the ideas we put forward in our Psych Review paper. Our model combined multiple Successor Representations which it switches between based on uncertainty. Jess's model adds reward outcomes to this process and captures splitter cells and more!

Andrew MacAskill @macaskillaf.bsky.social · May 29

Congrats to the fantastic Jess P for her new paper! She compared how feature Vs outcome focussed agents learn to solve contextual inference problems. She found that you need a balance of both to learn these tasks - and that this mix recapitulates pfc and hippocampal activity in rodent tasks!

bioRxiv Neuroscience @biorxiv-neursci.bsky.social · May 29

Contextual inference through flexible integration of environmental features and behavioural outcomes https://www.biorxiv.org/content/10.1101/2025.05.28.656607v1

1 1 4

Reposted by Jesse Geerts

Andrew MacAskill @macaskillaf.bsky.social · May 29

Congrats to the fantastic Jess P for her new paper! She compared how feature Vs outcome focussed agents learn to solve contextual inference problems. She found that you need a balance of both to learn these tasks - and that this mix recapitulates pfc and hippocampal activity in rodent tasks!

bioRxiv Neuroscience @biorxiv-neursci.bsky.social · May 29

Contextual inference through flexible integration of environmental features and behavioural outcomes https://www.biorxiv.org/content/10.1101/2025.05.28.656607v1

1 7 16

Jesse Geerts @jessegeerts.bsky.social · Jun 6

read the full paper here: arxiv.org/abs/2506.04289

Relational reasoning and inductive bias in transformers trained on a transitive inference task

Transformer-based models have demonstrated remarkable reasoning abilities, but the mechanisms underlying relational reasoning in different learning regimes remain poorly understood. In this work, we i...

arxiv.org

7

Jesse Geerts @jessegeerts.bsky.social · Jun 6

Many thanks to coauthors @scychan.bsky.social, @clopathlab.bsky.social, @neurokim.bsky.social !

1 5

Jesse Geerts @jessegeerts.bsky.social · Jun 6

The key insight: computational strategies underlying ICL aren't fixed but depend on both learning paradigm and pre-training structures. This helps explain when AI systems will generalize beyond their training data.

1 6

Jesse Geerts @jessegeerts.bsky.social · Jun 6

5. We could see these differences in their internal representations. Successful models organized items along continuous dimensions in representation space, while unsuccessful models showed no such structure.

1 4

Jesse Geerts @jessegeerts.bsky.social · Jun 6

4. Pre-training ICL models on linear regression tasks changed this outcome. These models then succeeded at transitive inference and didn't rely on induction circuits.

1 5

Jesse Geerts @jessegeerts.bsky.social · Jun 6

3. Mechanistic analysis revealed why: ICL models developed induction circuits - specialized attention patterns that implement match-and-copy operations rather than encoding hierarchical relationships.

1 6

Jesse Geerts @jessegeerts.bsky.social · Jun 6

2. In-context learning models failed to generalize transitively. Despite perfect performance on training pairs, they couldn't infer relationships between non-adjacent items.

1 5