Lightnews — Scholar-powered news

Reposted by Amir Ozhan Dehghani

Hafez Ghaemi @hafezghm.bsky.social · 22d

Excited to share that seq-JEPA has been accepted to NeurIPS 2025!

Hafez Ghaemi @hafezghm.bsky.social · May 14

Preprint Alert 🚀

Can we simultaneously learn transformation-invariant and transformation-equivariant representations with self-supervised learning?

TL;DR Yes! This is possible via simple predictive learning & architectural inductive biases – without extra loss terms and predictors!

🧵 (1/10)

2 2 13

Reposted by Amir Ozhan Dehghani

Pouya Bashivan 🇮🇷🇨🇦 @bashivan.bsky.social · Aug 9

We’ll be at CCN next week with two posters, one contributed talk and one satellite talk! If you are interested in neural coding in primate hippocampus, modeling visual search, or cortical topography come and talk to us! See👇for details:

1 4 20

Reposted by Amir Ozhan Dehghani

Shahab Bakhtiari @shahabbakht.bsky.social · Jul 31

A few years ago, with @tyrellturing.bsky.social @patrickmineault.bsky.social and others, we showed one single objective (prediction) was sufficient for explaining specialized pathways of mouse visual cortex: tinyurl.com/2vsjbve7

Super cool to see the same observation now for human visual cortex.

2 8 39

Amir Ozhan Dehghani @amirozhan.bsky.social · Apr 23

📌 Poster Session:
When📅: Friday, April 25 from 10:00 AM to 12:30 PM
Where: Hall 3 + Hall 2B #60
Check out our project page: amirozhan.github.io/CBSOM/

See you soon!

@mila-quebec.bsky.social

1 1

Amir Ozhan Dehghani @amirozhan.bsky.social · Apr 23

I’ll be at @iclr-conf.bsky.social presenting our work “Credit-based self organizing maps: training deep topographic networks with minimal performance degradation” — if you're into #NeuroAI, cortical topography, vision, or just up for a good convo, drop by and DM me! 🧠✨

1 1 8

Reposted by Amir Ozhan Dehghani

Shahab Bakhtiari @shahabbakht.bsky.social · Mar 14

📢 We have a new #NeuroAI postdoctoral position in the lab!

If you have a strong background in #NeuroAI or computational neuroscience, I’d love to hear from you.

(Repost please)

🧠📈🤖

2 40 59

Amir Ozhan Dehghani @amirozhan.bsky.social · Mar 9

People who might be interested: @taliakonkle.bsky.social, @mschrimpf.bsky.social, @apurvaratan.bsky.social @nancykanwisher.bsky.social

2

Amir Ozhan Dehghani @amirozhan.bsky.social · Mar 9

11/ Last (but not least), this work wouldn’t have been possible without the help of my collaborators
@Xinyu Qian, @asafarahani.bsky.social and research advisor @bashivan.bsky.social ⭐️

1 1

Amir Ozhan Dehghani @amirozhan.bsky.social · Mar 9

10/ Our results signify the potentially critical role of top-down assigned credits in shaping the topographical organization within the cortex.

1

Amir Ozhan Dehghani @amirozhan.bsky.social · Mar 9

9/ Finally, we show that compared to other baselines including classic SOM (AB-SOM) and TDANN, CB-SOM shows substantially higher alignment with neural activity in the visual cortex of both macaques (Brain-Score) and humans (NSD).

1

Amir Ozhan Dehghani @amirozhan.bsky.social · Mar 9

8/ Similarly, the model’s deeper layers replicated the organization of the higher-level visual cortical regions according to category-selectivity, resulting in category-selective filter clusters in these layers.

1

Amir Ozhan Dehghani @amirozhan.bsky.social · Mar 9

7/ We also observed an exponentially decaying pairwise correlation between filters as a function of their distance on the simulated cortical sheet.

1

Amir Ozhan Dehghani @amirozhan.bsky.social · Mar 9

6/ Our model replicated the topographical organization seen in the primary visual cortex topographical organization, including smooth transitions in selectivity according to stimulus orientation and spatial frequency.

1

Amir Ozhan Dehghani @amirozhan.bsky.social · Mar 9

5/ With this approach, our ResNet-18 model achieved a substantial performance boost while preserving topographical organization.

1

Amir Ozhan Dehghani @amirozhan.bsky.social · Mar 9

4/ We proposed an alternative approach based on self-organizing maps, Credit-Based Self-Organizing Maps(CB-SOMs), where the competition mechanism is guided by the importance of the assigned credit to each weight – i.e. filters with the highest gradient norm in each conv layer.

1

Amir Ozhan Dehghani @amirozhan.bsky.social · Mar 9

3/ Previous approaches to simulate topographical organization in neural networks involved using various objective functions, but this often results in weaker performance on object categorization.
cell.com/neuron/fullt...
pnas.org/doi/10.1073/...
arxiv.org/pdf/2308.09431

https://cell.com/neuron/fulltext/S0896-6273(24)00279-4…

1 1

Amir Ozhan Dehghani @amirozhan.bsky.social · Mar 9

2/ Cortex in many animals, including humans, is topographically organized, where functionally similar neurons are spatially close to each other. We sought to understand the computational principles underlying this visual topographical organization.

1

Amir Ozhan Dehghani @amirozhan.bsky.social · Mar 9

I am excited to share that my undergraduate research has now been accepted to #ICLR2025.
📎: openreview.net/pdf?id=wMgr7...
🧑‍💻: github.com/BashivanLab/...

Tweeprint🧵⬇️

1 4 17

Amir Ozhan Dehghani @amirozhan.bsky.social · Mar 9

"Hello World"