Lightnews — Scholar-powered news

Fatemeh_Hadaeghi @fatemehhadaeghi.bsky.social · 2d

A thought-provoking perspective from the visionary @giacomoi.bsky.social, calling for neuromorphic computing to return to its root: fundamental neuroscience; an inspiring vision for the future of NeuroAI 🤩

Neuron @cp-neuron.bsky.social · 5d

Neuromorphic is dead. Long live neuromorphic

In this NeuroView, Giacomo Indiveri discusses the origins and evolution of neuromorphic engineering, reflects on recent shifts toward digital and AI-centric approaches, and advocates for a revival of the field’s original principles, emphasizing analog, brain-inspired systems and the importance of fundamental neuroscience for future progress.

dlvr.it

4 12

Fatemeh_Hadaeghi @fatemehhadaeghi.bsky.social · 11d

I would also like to thank prominent figures in the field, Sara Solla, Petra Vertes, @kenmiller.bsky.social, @bendfulcher.bsky.social, @jlizier.bsky.social, @danakarca.bsky.social, MarcusKaiser, Gorka Zamora-López, and Patrick Desrosiers, who provided feedback during lab visits and conferences.

4

Fatemeh_Hadaeghi @fatemehhadaeghi.bsky.social · 11d

This work has been in progress for 3+ years.
Grateful to my co-authors, Claus Hilgetag, @kayson.bsky.social, and Moein Khajehnejad for their invaluable contributions,

1 4

Fatemeh_Hadaeghi @fatemehhadaeghi.bsky.social · 11d

Implications:
🧠 Neuroscience — functional rationale for the evolutionary suppression of strong reciprocal motifs.
🤖 NeuroAI — reciprocity as a tunable design parameter in recurrent & neuromorphic networks.

1 3

Fatemeh_Hadaeghi @fatemehhadaeghi.bsky.social · 11d

So why does the brain avoid strong loops?
Because reciprocity systematically hurts computation.

Suppressing strong loops preserves:
working memory
representational diversity
stable but flexible dynamics

1 3

Fatemeh_Hadaeghi @fatemehhadaeghi.bsky.social · 11d

We validated this on empirical connectomes (macaque long-distance, macaque visual cortex, marmoset).

Result: the same pattern.
Strong reciprocity consistently undermines memory and representational richness.

1 2

Fatemeh_Hadaeghi @fatemehhadaeghi.bsky.social · 11d

Spectral analysis explains why:

- Higher reciprocity → larger spectral radius (instability risk).
- Narrower spectral gap → less dynamical diversity.
- Lower non-normality → weaker transient amplification.

Together → compressed dynamical range.

1 3

Fatemeh_Hadaeghi @fatemehhadaeghi.bsky.social · 11d

Interestingly, hierarchical modular networks consistently outperformed random counterparts, but only when reciprocity was low. However, the comparative advantages of network topologies shift with reciprocity, sparsity, and weight distribution

1 3

Fatemeh_Hadaeghi @fatemehhadaeghi.bsky.social · 11d

Findings (robust across sizes, densities, architectures):
- Increasing reciprocity (link as well as strength reciprocity) reduces memory capacity.
- Representation becomes less diverse (lower kernel rank).
- Effects are strongest in ultra-sparse networks.

1 3

Fatemeh_Hadaeghi @fatemehhadaeghi.bsky.social · 11d

Methods:
- Reservoir computing to isolate structure from learning.
- Networks of 64–256 nodes, in both ultra-sparse and sparse regimes.
- Topologies: small-world, hierarchical modular, core–periphery, hybrid, and nulls.
- Metrics: memory capacity, kernel rank, spectral analysis.

1 2

Fatemeh_Hadaeghi @fatemehhadaeghi.bsky.social · 11d

In earlier work (www.biorxiv.org/content/10.1...), we developed Network Reciprocity Control (NRC): algorithms that adjust reciprocity (link + strength) while preserving network structure.

In this study, we apply NRC to systematically test how reciprocity shapes computation in recurrent networks.

1 2

Fatemeh_Hadaeghi @fatemehhadaeghi.bsky.social · 11d

The no-strong-loops principle:
Across species (macaque, marmoset, mouse), strong reciprocal (symmetric) connections are rare.

This asymmetry is well known anatomically.
But what are its computational consequences?

1 2

Fatemeh_Hadaeghi @fatemehhadaeghi.bsky.social · 11d

🚨 New preprint!
“A Computational Perspective on the No-Strong-Loops Principle in Brain Networks”
www.biorxiv.org/content/10.1...

Over the past 3 years, we’ve been investigating why cortical networks avoid strong reciprocal loops — and what this means for computation.

1 4 16

Reposted by Fatemeh_Hadaeghi

Alex Fornito @alexfornito.bsky.social · Aug 17

Interested in Network hubs, cortical hierarchies, and gradients? Ever wonder where they come from? Check our latest review, where we cover different approaches to mapping hubs, models for their evolution, and mechanisms for how they develop:

osf.io/preprints/os...

1 40 97

Reposted by Fatemeh_Hadaeghi

Dr Dionne Argyropoulos @dionnecargy.bsky.social · Jul 11

Have you ever been transfixed by the colour palette of the Melbourne tram network?

Well I have... so I made a package called {melbourne}! So far it includes a colour palette called "melb_trams()". More to come, stay tuned!

🔗 Check it out: github.com/dionnecargy/...

#RStats #DataScience #Rcoding

GitHub - dionnecargy/melbourne: A package love letter to Melbourne

A package love letter to Melbourne. Contribute to dionnecargy/melbourne development by creating an account on GitHub.

github.com

5 21

Reposted by Fatemeh_Hadaeghi

Kayson Fakhar @kayson.bsky.social · Jul 6

Good morning folks. If you’re around #OCNS2025, maybe come by today for a chat about optimal communication in brain networks? ✨

10 30

Fatemeh_Hadaeghi @fatemehhadaeghi.bsky.social · Jun 26

Wonderful work by Shrey and Kayson in developing an XAI method for exploring the contribution of any computational unit (e.g., nodes, experts, communities, filters) within neural networks. It enables analysis of both their influence on each other and their overall impact on task performance.

Shrey Dixit @shreydixit.bsky.social · Jun 25

🚨Preprint Alert: Who Does What in Deep Learning? Multidimensional Contribution of Neural Units using Game Theory: arxiv.org/abs/2506.19732
The result of my MSc thesis is out with @kayson.bsky.social @fatemehhadaeghi.bsky.social @patrickmineault.bsky.social @kordinglab.bsky.social, Claus C. Hilgetag

Who Does What in Deep Learning? Multidimensional Game-Theoretic Attribution of Function of Neural Units

Neural networks now generate text, images, and speech with billions of parameters, producing a need to know how each neural unit contributes to these high-dimensional outputs. Existing explainable-AI ...

arxiv.org

1 6

Reposted by Fatemeh_Hadaeghi

Shrey Dixit @shreydixit.bsky.social · Jun 25

🚨Preprint Alert: Who Does What in Deep Learning? Multidimensional Contribution of Neural Units using Game Theory: arxiv.org/abs/2506.19732
The result of my MSc thesis is out with @kayson.bsky.social @fatemehhadaeghi.bsky.social @patrickmineault.bsky.social @kordinglab.bsky.social, Claus C. Hilgetag

Who Does What in Deep Learning? Multidimensional Game-Theoretic Attribution of Function of Neural Units

Neural networks now generate text, images, and speech with billions of parameters, producing a need to know how each neural unit contributes to these high-dimensional outputs. Existing explainable-AI ...

arxiv.org

1 13 27

Reposted by Fatemeh_Hadaeghi

Tatia Buidze @tatiabu.bsky.social · Jun 18

✨Excited to share that our new paper is now out in iScience!✨

🧠 We show that people can coordinate surprisingly well in novel interactions by violating others' expectations - without requiring deep, recursive reasoning about others’ beliefs.

📄 Read the full paper here: www.cell.com/iscience/ful...

Expectation violations as an effective alternative to complex mentalizing in novel communication

Neuroscience; Systems neuroscience; Social sciences

www.cell.com

11 27

Reposted by Fatemeh_Hadaeghi

Marcus Ghosh @marcusghosh.bsky.social · Jun 11

How should #multisensory signals be combined when they are structured in time?

To explore this, we:
* Introduce a new multisensory task
* Compare several models

TLDR:
* Prior models perform suboptimally
* Our new model performs ≈ an RNN, while using less than 1/10th the number of parameters.

Dan Goodman @neural-reckoning.org · Jun 10

Fusing multisensory signals across channels and time.

Now published at PLOS Comp Biol! 🎉 With @swathianil.bsky.social and @marcusghosh.bsky.social.

journals.plos.org/ploscompbiol...

TLDR, when multisensory signals vary over time, neural architecture becomes important. Biggest not always best.

3 17

Reposted by Fatemeh_Hadaeghi

Matteo Saponati @matteosaponati.bsky.social · Feb 18

1/ I am very excited to announce that our paper "The underlying structures of self-attention: symmetry, directionality, and emergent dynamics in Transformer training" is available on arXiv 💜

arxiv.org/abs/2502.10927

How is information encoded in self-attention matrices?How to interpret it?

⬇️

The underlying structures of self-attention: symmetry, directionality, and emergent dynamics in Transformer training

Self-attention is essential to Transformer architectures, yet how information is embedded in the self-attention matrices and how different objective functions impact this process remains unclear. We p...

arxiv.org

1 3 8

Reposted by Fatemeh_Hadaeghi

Daniele Marinazzo @danielemarinazzo.bsky.social · May 15

PhD fellowship to work with me and Benedetta Franceschiello on the analysis and modelling of fast sampled fMRI data!

www.ugent.be/en/work/scie...

Doctoral fellow

www.ugent.be

10 11

Reposted by Fatemeh_Hadaeghi

Nature Neuroscience @natneuro.nature.com · Apr 11

A Perspective on integrating multimodal data to understand cortical circuit architecture and function

@alleninstitute.bsky.social

www.nature.com/articles/s41...

Integrating multimodal data to understand cortical circuit architecture and function - Nature Neuroscience

This paper discusses how experimental and computational studies integrating multimodal data, such as RNA expression, connectivity and neural activity, are advancing our understanding of the architectu...

www.nature.com

3 19

Reposted by Fatemeh_Hadaeghi

Mareike Gann @mareikegann.bsky.social · Apr 4

Excited to share our latest preprint on connectivity modulation with dual-site tACS! We show that in-phase tACS at 20 Hz can disrupt fMRI connectivity between the primary motor cortices, but also affects connectivity with other motor regions.

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

5 6

Fatemeh_Hadaeghi @fatemehhadaeghi.bsky.social · Apr 3

Found a #bifurcation (bifurcated ?) #tree outside the Royal Botanic garden, #Sydney 😬

2