shreydixit.bsky.social
Shrey Dixit
@shreydixit.bsky.social
47 followers 82 following 17 posts
Doctoral Researcher doing NeuroAI at the Max Planck Institute of Human Cognitive and Brain Sciences
Posts Media Videos Starter Packs
Pinned
shreydixit.bsky.social
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 Shrey Dixit
andrejbicanski.bsky.social
Andrej Bicanski @andrejbicanski.bsky.social · 6d
New preprint from the lab and great work by Fei Wang. We show how subiculum trace vector cells can be modeled consistent with known effects in CA1. Traces are driven by a mismatch learning rule to keep associative memories in line with experience.
www.biorxiv.org/content/10.1...
Dynamic updating of cognitive maps via traces of experience in the subiculum
In the classical view of hippocampal function, the subiculum is assigned the role as the output layer. In spatial paradigms, some subiculum neurons manifest as so-called boundary vector cells (BVCs), ...
www.biorxiv.org
1 10 19
Reposted by Shrey Dixit
qiaoli-huang.bsky.social
Qiaoli Huang @qiaoli-huang.bsky.social · Aug 26
Happy to share our new preprint. We found that brain adaptively switches between relational and item-based coding strategies based on the structure of memory contents.
@doellerlab.bsky.social

www.biorxiv.org/content/10.1...
Efficient coding in working memory is adapted to the structure of the environment
Working memory (WM) relies on efficient coding strategies to overcome its limited capacity, yet how the brain adaptively organizes WM representations to maximize coding efficiency based on environment...
www.biorxiv.org
1 10 36
Reposted by Shrey Dixit
nichome.bsky.social
Nicholas Menghi @nichome.bsky.social · 15d
🚨 New preprint! Impact of Task Similarity and Training Regimes on Cognitive Transfer and Interference 🧠

We compare humans and neural networks in a learning task, showing how training regime and task similarity interact to drive transfer or interference.

www.biorxiv.org/content/10.1...
Impact of Task Similarity and Training Regimes on Cognitive Transfer and Interference
Learning depends not only on the content of what we learn, but also on how we learn and on how experiences are structured over time. To investigate how task similarity and training regime interact dur...
www.biorxiv.org
1 7 16
Reposted by Shrey Dixit
doellerlab.bsky.social
Doeller Lab @doellerlab.bsky.social · 15d
Join us at the Max Planck Institute in Leipzig as a Postdoc to explore cognitive maps in the human brain: learning, memory & the formation of structural representations. Excellent infrastructure with a leading scientific network. Apply by 13 October: postdocprogram.mpg.de/node/21187
2 3
Reposted by Shrey Dixit
matthiasmichel.bsky.social
Matthias Michel @matthiasmichel.bsky.social · Apr 21
Very happy to announce that our paper “Sensory Horizons and the Functions of Conscious Vision” is now out as a target article in BBS!! @smfleming.bsky.social and I present a new theory of the evolution and functions of visual consciousness. Article here: doi.org/10.1017/S014.... A (long) thread 🧵
Sensory Horizons and the Functions of Conscious Vision | Behavioral and Brain Sciences | Cambridge Core
Sensory Horizons and the Functions of Conscious Vision
doi.org
6 69 180
Reposted by Shrey Dixit
marcusghosh.bsky.social
Marcus Ghosh @marcusghosh.bsky.social · Aug 1
How does the structure of a neural circuit shape its function?

@neuralreckoning.bsky.social & I explore this in our new preprint:

doi.org/10.1101/2025...

🤖🧠🧪

🧵1/9
A diagram showing 128 neural network architectures.
5 35 98
shreydixit.bsky.social
Shrey Dixit @shreydixit.bsky.social · Jul 31
Finally, huge thanks to the organizers for giving us this opportunity @algonautsproject.bsky.social
1
shreydixit.bsky.social
Shrey Dixit @shreydixit.bsky.social · Jul 31
What's next? We plan to publish a more in-depth analysis of VIBE’s internal dynamics and feature–parcel mappings, to advance our understanding of brain function, and guide future work in neuroscience.
Spoiler: We already have a model that beats the top score of this challenge ;)
1 1
shreydixit.bsky.social
Shrey Dixit @shreydixit.bsky.social · Jul 31
Shapley (MSA) Insights:
Feature attribution maps align with neuroanatomy, but crazily enough, textual features from the transcripts are the most predictive of all.
1 1
shreydixit.bsky.social
Shrey Dixit @shreydixit.bsky.social · Jul 31
We have strong lifts over baseline both in-distribution and OOD. VIBE (final): r=0.3225 (ID, Friends S07) & 0.2125 (6×OOD).
Competition submission (earlier iteration): r=0.3198 ID / 0.2096 OOD → 1st in Phase-1, 2nd overall.
vs baseline 0.2033/0.0895 → +0.119/+0.123.
1 1
shreydixit.bsky.social
Shrey Dixit @shreydixit.bsky.social · Jul 31
We present VIBE: Video-Input Brain Encoder, which is a 2-stage Transformer. First stage merges text, audio, and visual features per timepoint (plus subject embeddings). And the second stage models temporal dynamics with rotary positional embeddings.
1 1
shreydixit.bsky.social
Shrey Dixit @shreydixit.bsky.social · Jul 31
Competition & Data: Algonauts 2025 tests how well we can predict brain activity while people watch naturalistic movies. Multi-modal stimuli (video, audio, text) → whole-brain fMRI, split into parcels. Train on movies/TV; evaluate in-distribution and on out-of-distribution films.
1 1
shreydixit.bsky.social
Shrey Dixit @shreydixit.bsky.social · Jul 31
We did it! 🏆 We won Phase 1 and placed 2nd overall in the Algonauts 2025 Challenge. So proud of the crew
@keckjanis.bsky.social,Viktor Studenyak,Daniel Schad,Aleksandr Shpilevoi. Huge thanks to @andrejbicanski.bsky.social and @doellerlab.bsky.social for support. Report: arxiv.org/abs/2507.17958
1 3 10
shreydixit.bsky.social
Shrey Dixit @shreydixit.bsky.social · Jul 29
I personally prefer subway surfers. This one couldn't hold my attention for the whole video
2
Reposted by Shrey Dixit
jbimaknee.bsky.social
Brad Aimone @jbimaknee.bsky.social · Jul 29
For a few years I have said that neuromorphic is specialized general purpose, like GPUs, but with different advantages.

In this preprint I try to put some substance to that claim. There are real theoretical advantages, but they aren't obvious. 🧪🧠🤖 www.arxiv.org/abs/2507.17886
Neuromorphic Computing: A Theoretical Framework for Time, Space, and Energy Scaling
Neuromorphic computing (NMC) is increasingly viewed as a low-power alternative to conventional von Neumann architectures such as central processing units (CPUs) and graphics processing units (GPUs), h...
www.arxiv.org
2 6 38
shreydixit.bsky.social
Shrey Dixit @shreydixit.bsky.social · Jul 24
Aside from a few mispronunciations, the AI really got the paper. Honestly, it's reassuring—if an AI can follow it without any mistakes, then folks in the field probably can too :)
1
shreydixit.bsky.social
Shrey Dixit @shreydixit.bsky.social · Jul 24
Just came across an AI-generated video summary/review of our recent preprint—and I have to say, I’m genuinely impressed. It does a great job summarizing the paper, and I’d actually recommend it to others. Check it out: www.youtube.com/watch?v=3g5K...
Multidimensional Game-Theoretic Attribution of Function of Neural Units
YouTube video by LuxaK
www.youtube.com
1 3
Reposted by Shrey Dixit
ianholmes.org
Ian Holmes @ianholmes.org · Jul 5
White text on white background instructing LLMs to give positive reviews is apparently now common enough to show up in searches for boilerplate text.
neuralnoise.com
Pasquale Minervini @neuralnoise.com · Jul 5
"in 2025 we will have flying cars" 😂😂😂
9 220 530
Reposted by Shrey Dixit
kayson.bsky.social
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
Reposted by Shrey Dixit
marcelomattar.bsky.social
Marcelo Mattar @marcelomattar.bsky.social · Jul 2
Thrilled to see our TinyRNN paper in @nature! We show how tiny RNNs predict choices of individual subjects accurately while staying fully interpretable. This approach can transform how we model cognitive processes in both healthy and disordered decisions. doi.org/10.1038/s415...
Discovering cognitive strategies with tiny recurrent neural networks - Nature
Modelling biological decision-making with tiny recurrent neural networks enables more accurate predictions of animal choices than classical cognitive models and offers insights into the underlying cog...
doi.org
9 140 330
Reposted by Shrey Dixit
keckjanis.bsky.social
Janis Keck @keckjanis.bsky.social · Jun 25
Excited to share that our work on the interplay of symmetric learning rules and successor representations (SR) is published in @plos.org (comp.bio)

Work done together with @doellerlab.bsky.social, @caswell.bsky.social and Juergen Jost.

doi.org/10.1371/jour...
#neuroskyence #compneurosky

1/n
Impact of symmetry in local learning rules on predictive neural representations and generalization in spatial navigation
Author summary The hippocampus is a brain region which plays a crucial role in spatial navigation for both animals and humans. Contemporarily, it’s thought to store predictive representations of the e...
doi.org
1 8 23
shreydixit.bsky.social
Shrey Dixit @shreydixit.bsky.social · Jun 25
Not really. Thankfully, Max Planck has GPU clusters that I can use.
Although I did ask my friend (o3) about it, according to whom, 5090 is sufficient for most cases. (chatgpt.com/share/685c6a...)
ChatGPT - RTX 5000 vs 5090 for ML
Shared via ChatGPT
chatgpt.com
2
shreydixit.bsky.social
Shrey Dixit @shreydixit.bsky.social · Jun 25
DCGAN Case Study:
Pixel-wise Shapley Modes reveal the inverted CNN hierarchy: first transposed-conv layer shapes high-level facial parts; final layer merely renders RGB channels.
3
shreydixit.bsky.social
Shrey Dixit @shreydixit.bsky.social · Jun 25
LLM Case Study:
Calculated expert-level contributions of an MOE-based LLM across arithmetic, language ID, and factual recall. Found an expert which was super-important for all domains. Also found redundant experts, removing which does not decrease performance much.
1 3
shreydixit.bsky.social
Shrey Dixit @shreydixit.bsky.social · Jun 25
MLP Case Study:
Neural computations within a three-layer MNIST MLP were analysed. L1/L2 regularisation funnels computations into a few neurons. Also, contrary to popular belief, large weights do not equal high importance of neural units.
1 3