Pouya Bashivan 🇮🇷🇨🇦

Local lateral connectivity is sufficient for replicating cortex-like topographical organization in deep neural networks Across the primate cortex, neurons that perform similar functions tend to be spatially grouped together. In the high-level visual cortex, this biological principle manifests itself as a modular organi...

You might this relevant too, in terms of robustness

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

Are you thinking of automating the review process? LLMs are already doing some of that. I’m pretty sure at least 25% of reviews at conferences are already LLM-generated.

I'm not sure which two papers you are talking about but I didn't notice the PNAS paper I posted in the preprint

interesting

Went over the Centaur paper in lab meeting today. I got stuck on this plot, thinking initially the small difference between Llama and Centaur is bothering me. But the real issue hit me an hour later... Each layer's neural similarity is with the **whole** brain! No single layer is expected to sim 🧠

Congratulations on the paper! It made me think of this paper by Pawan Sinha on developmentally aligned training of neural networks: www.pnas.org/doi/full/10....
It's interesting to see that this procedure leads to much broader robustness than previously thought.

Grant reviewer on my proposal for using DNNs to model working memory: “The ability to yield new information is very limited because WM circuits have already been abundantly specified in clinical and preclinical studies.” So, it’s entirely solved now, eh? How long did I sleep last night?

Huge congratulations to @silicolabs.bsky.social on this achievement.

Excited about this collaboration on visuomotor learning in VR.

We also have some funding available for graduate students interested in doing research in this area. Reach out if you're interested.

🧠📈 🧠🤖 #VisionScience

Context-dependent computation by recurrent dynamics in prefrontal cortex - Nature This study shows that in monkeys making context-dependent decisions, task-relevant and task-irrelevant signals are confusingly intermixed in single units of the prefrontal cortex, but are readily unde...

Attractors are great for framing neural dynamics, but I agree, by themselves they’re **not** mechanisms. My favourite papers don’t just point to attractors; they show how those attractors drive tasks (e.g., the 2013 @valeriomante & @sussillodavid.bsky.social classic: nature.com/articles/nature12742)

Self supervised learning of spatial representations from episodic memories. Led by undergraduate student in my lab. Very proud!

Congratulations Dan! I’m sure you’ll do great things in your lab.

Announcement: Workshop at #CCN2025
🧠 Modeling the Physical Brain: Spatial Organization & Biophysical Constraints
🗓️ Monday, Aug 11 | 🕦 11:30–18:00 CET | 📍 Room A2.07
🔗 Register: tinyurl.com/CCN-physical...
#NeuroAI @cogcompneuro.bsky.social

How well do models of visual cortex generalize to out of distribution samples? Author summary Inspired by the neural circuits of the brain, deep neural networks (DNN) have been steadily improving in their ability to perform foundational visual tasks such as object recognition. W...

Nice work! We also showed recently that performances on typical benchmarks are not good measures of neural predictivity in macaques anymore. Improving robustness could still boost predicivitity a bit though

journals.plos.org/ploscompbiol...

If you are at #ICLR2025 and interested in topographic neural networks, come by @amirozhan.bsky.social 's poster. See the info below!

The study of plasticity has always been about gradients Abstract figure legend This figure illustrates our central point in this paper using a well-known meme. That central point is that the physiological study of synaptic plasticity and learning has alwa...

A point of pride for me was publishing a paper with a meme as a graphical abstract in a journal that dates back to 1878:

physoc.onlinelibrary.wiley.com/doi/full/10....

This is apparently a #TopViewedArticle, so the achievement got the recognition it deserves! 😅

CC @kordinglab.bsky.social

Don’t let Poilievre crush your CBC crush!

From Paul Gross to Peter Mansbridge, Catherine O’Hara to the Levys—CBC has brought us icons we’ve loved for decades.

Lose CBC, lose the crushes.

#SaveTheCBC #CBCrushed #OurVoiceOurCBC #CdnPoli #PublicBroadcasting

Renowned neuroscientist Melanie Woodin has been named #UofT’s 17th president — and asks the community to join in her in “dreaming big.” #UofT alumna Woodin starts her five-year term July 1st. ➡️ uoft.me/uoftpres17

At #COSYNE2025? Come by Motahareh's poster at 1pm [2-089] if you like to know about how cool her NN model of visual search is! Trained de novo on visual search, its fast, efficient, mimics human behaviour, and has a beautiful neural geometry.

Check out our new paper on making deep networks topographical with a new variant of SOMs! Very proud of @amirozhan.bsky.social who did this work as an undergrad!

Cosyne 2025 undergraduate mentor applications The Cosyne Undergraduate Travel Grant, is a program that brings ~12 undergraduates from across the world to Cosyne. As part of the program, we aim to pair the students with a “conference mentor”, a P...

Recruiting phd/postdocs to mentor Undergrad Awardees!!

This program brings UGs from across the 🌎 considering a career in neuro research

Duties: meet students on 1st day, take them to a poster session, help them feel welcome.

Apply: (very short!) form
forms.gle/9hCniCA29ky5...

If you are a neuroscientist, please use SFN's site to write to your reps about the importance of funding for science (they have letters ready to go, you can just use theirs or write/edit as you see fit): www.sfn.org/advocacy/adv...

That says how little I know about meme contests 😊

I’m curious what the students’ guesses were

Pretty cool, Ratan and team!

Neurogenesis event visual: Speakers Paul Masset, PhD, and Pouya Bashivan, PhD February 5, 2025, from 4:30-5:30 p.m. (with post-event reception)

Headshot of Paul Masset and Pouya Bashivan

🌟 Sign up for the Neurogenesis Speaker Series!

Join us on Feb 5 at @theneuro.bsky.social for talks featuring two HBHL-supported faculty, @paulmasset.bsky.social & @bashivan.bsky.social, as they share their groundbreaking research in brain health at @mcgill.ca. 🎙️

🔗 Register today: mcgill.ca/x/wHe

Geometry of naturalistic object representations in recurrent neural network models of working memory Working memory is a central cognitive ability crucial for intelligent decision-making. Recent experimental and computational work studying working memory has primarily used categorical (i.e., one-hot)...

We came up with our own orthogonality index when axes across two spaces can be paired. Eq 1 in this paper:

arxiv.org/abs/2411.02685

Fooling LLM graders into giving better grades through neural activity guided adversarial prompting The deployment of artificial intelligence (AI) in critical decision-making and evaluation processes raises concerns about inherent biases that malicious actors could exploit to distort decision outcom...

Our new paper "Fooling LLM graders into giving better grades through neural activity guided adversarial prompting" lead expertly by Atsushi Yamamura shows how even in closed models like Gemini, we can add a small adversarial suffix to an essay to get unreasonably high scores
arxiv.org/abs/2412.15275

Stonehenge Reports

If you’re at NeurIPS, Check out @xiaoxuanlei.bsky.social poster on representational geometry in RNN models of working memory!

Different paths toward safe AI at different Marr's levels

Excited to release what we’ve been working on at Amaranth Foundation, our latest whitepaper, NeuroAI for AI safety! A detailed, ambitious roadmap for how neuroscience research can help build safer AI systems while accelerating both virtual neuroscience and neurotech. 1/N