Brains and Computers @GoogleDeepMind + @GeorgiaTech
🛸
Joined September 2019
- Tweets223
- Following429
- Followers523
- Likes10.3K
Life update: I've joined @GoogleDeepMind to work with @joaocarreira and team on video models!
inspired by @kellerjordan0's nanoGPT speedruns, I made druta (द्रुत) - a fast video dataset format for #PyTorch for when storage is cheap but time is not
orders of magnitude faster than decord btw
pip install druta
github.com/Mayukhdeb/druta
mayukh09 retweeted
no wrestling with obscure AWS cli commands,
no massive figshare zip files,
zero pain fMRI dataset that takes <5 min to get running
pip install mosaic-dataset
🧠Excited to share MOSAIC: a unified fMRI mega-dataset: 430k+ trials, 160k stimuli, 93 subjects (and more to come!) all in a single brain space
Enables massive scale for modeling the visual cortex with just 2 lines of code
github.com/murtylab/mosaic-d…
🐍pip install mosaic-dataset
🧠Excited to share MOSAIC: a unified fMRI mega-dataset: 430k+ trials, 160k stimuli, 93 subjects (and more to come!) all in a single brain space
Enables massive scale for modeling the visual cortex with just 2 lines of code
github.com/murtylab/mosaic-d…
🐍pip install mosaic-dataset
🌐Check out the website: blahner.github.io/MOSAICfmri…
And our preprint👀: biorxiv.org/content/10.64898…
mayukh09 retweeted
🚨 Paper alert:
To appear in the DBM Neurips Workshop
LITcoder: A General-Purpose Library for Building and Comparing Encoding Models
📄 arxiv: arxiv.org/abs/2509.09152v1
🔗 project: litcoder-brain.github.io/
Come check out our Spotlight paper - TopoNets at #ICLR2025!
DM me if you're interested in discussing structured LLMs, vision models and brain-inspired algorithms in Singapore!
24 Apr 3:00pm-5:30pm - Hall 3 + Hall 2B #58
Website: toponets.github.io
My work got featured in @TEDAIVienna👀
AtMan (NeurIPS 2023) explains LLM outputs, telling you which part of the context was critical.
ted.com/talks/jonas_andrulis…
neurips.cc/virtual/2023/post…
mayukh09 retweeted
I always love to see brain topology inspired work. The brain saves so much on compute with useful topology/wiring already between retina and the first layer of the visual cortex. We should not forget where neural networks came from and should keep imitating nature. Great work!
Very proud to share our @iclr_conf paper: TopoNets! High-performing vision and language models with brain-like topography! Expertly led by grad student @mayukh091 and @MainakDeb19! A brief thread..
Watch how the science-selective neurons self-organize in a layer when I train a GPT w/ topoloss toponets.github.io
This figure could not make it to the paper but I liked it
mayukh09 retweeted
signal boosting this because it is both intuitive and extremely cool
Very proud to share our @iclr_conf paper: TopoNets! High-performing vision and language models with brain-like topography! Expertly led by grad student @mayukh091 and @MainakDeb19! A brief thread..
mayukh09 retweeted
Very proud to share our @iclr_conf paper: TopoNets! High-performing vision and language models with brain-like topography! Expertly led by grad student @mayukh091 and @MainakDeb19! A brief thread..
mayukh09 retweeted
Unlike the brain, most ANNs lack any kind of organization of units🧠
In our paper (ICLR 2025) with @mayukh091 @apurvaratan, we present TopoLoss, a simple, scalable way to add brain-like topography to AI models with minimal performance loss🧵
toponets.github.io
TopoNets is accepted into ICLR 2025 🥳
Inducing brain-like self-similarity in the weights of vision and language models yields:
- lower dimensionality
- parameter efficiency
- brain-like functional organization
toponets.github.io
Work done with @MainakDeb19 and @apurvaratan
Also huge thanks to @neuranna, @NeuroTaha and everyone at MurtyLab murtylab.com 🫡
mayukh09 retweeted
🚀At @NeurIPSConf tomorrow? Don't miss @nikolasmcneal and @MainakDeb19 's poster at the @unireps workshop on the adversarial sensitivity of vision encoding models of fMRI responses! A brief teaser about what they find.
If you make a drawing in the weight matrices of your neural network at initialization, it will likely still be visible at the end of training arxiv.org/abs/2012.02550
