Paul Scotti (@humanscotti) / X

Paul Scotti

586 posts

Paul Scotti

@humanscotti

CTO / Co-Founder @SophontAI @MedARC_AI | Visiting research scientist @PrincetonNeuro | ex @StabilityAI

Oakland, CA

Joined July 2009

Pinned
Paul Scotti
@humanscotti
Mar 19, 2024
🧠👁️Our MindEye2 preprint is out! We reconstruct seen images from fMRI brain activity using only 1 hour of training data. This is possible by first pretraining a shared-subject model using other people's data, then fine-tuning on a held-out subject with only 1 hour of data.🧵
104K
Paul Scotti
@humanscotti
May 30, 2023
Announcing 🧠👁️ MindEye! Our state-of-the-art fMRI-to-image approach that retrieves & reconstructs images from brain activity, is out as a preprint! MindEye takes human brain activity as input and outputs reconstructed images like these. Project page: medarc-ai.github.io/mindeye/
90K
Paul Scotti
@humanscotti
Feb 9, 2025
I resigned from Stability AI after over a year working as Head of Neuroimaging & AI. I’m now looking into new ways to support open-source science and conduct research on neuroAI and medical AI more broadly. A personal retrospective below...
22K
Paul Scotti
@humanscotti
Jun 17, 2024
Reconstructing seen images from brain activity using fMRI brain activity downsampled to portable fNIRS resolution (from 2mm voxels to 1cm voxels!) @_marleyx & @raffi_hotter organized a crazy hackathon where we tested this and designed & prototyped an fNIRS device from scratch
29K
Paul Scotti
@humanscotti
Jul 12, 2024
New dataset of 175 hours of EEG data from a single subject Authors found increasing dataset size leads to increasing improvement in classification of 5-second audio segments suggests performance can continue to increase with more data (scaling law)
arxiv.org
Scaling Law in Neural Data: Non-Invasive Speech Decoding with 175...
Brain-computer interfaces (BCIs) hold great potential for aiding individuals with speech impairments. Utilizing electroencephalography (EEG) to decode speech is particularly promising due to its...
18K
Paul Scotti
@humanscotti
Nov 14, 2023
Happy to share that I am now Head of Neuroimaging at @StabilityAI! My role entails leading the @MedARC_AI Neuroimaging & AI Lab: medarc.ai/fmri This lab at MedARC changes the usual way research labs work... 🧵 1/n
72K
Paul Scotti
@humanscotti
Sep 17, 2024
Nicole mentally imagined images from her past inside an MRI machine, and we used neuroAI to reconstruct those memories from her brain scans. My collab w/ Fujifilm connecting science & art to visualize the intangible aspects of memory and thought!
20K
Paul Scotti
@humanscotti
Dec 29, 2020
Replying to @ash1eyruba and @ashleyruba
That's awesome and very impressive! i want to add: some never get published, for reasons out of their control (bc of unclear data, mental health issues, bad work environment) and they and their work matter too. Unfortunately staying in academia can be a lot harder in these cases.
Paul Scotti
@humanscotti
Mar 31, 2025
AI for healthcare right now consists of domain-specific opaque models that fall prey to the parable of the blind men and the elephant Blind men = unimodal medical models Elephant = patient An untried niche in med AI infrastructure is to build open-source multimodal AI
9.1K
Paul Scotti
@humanscotti
Jan 26, 2022
Very happy to share that in April I will begin a postdoc at @PrincetonNeuro with @ptoncompmemlab to work on projects involving real-time fMRI! (Have already been familiarizing myself with rt-cloud which simplifies the complexities of setting up real-time, preprint below)
Norman Lab
@ptoncompmemlab
Jan 26, 2022
New preprint! RT-Cloud: A Cloud-based Software Framework to Simplify and Standardize Real-Time fMRI osf.io/sbrg7/ 🧵
Paul Scotti
@humanscotti
Oct 26, 2024
NeuroAI is a new field where trainees often feel caught between AI and neuro groups, lacking dedicated programs. Our new Nature Comms piece explores this, offering survey results & resources for trainees and institutions. Thanks to @loopyluppi & @achterbrain for leading this!
Nature Communications
@NatureComms
Oct 24, 2024
New developing area of #NeuroAI at the intersection of #neuroscience and #artificialintelligence has many open challenges. Here, @loopyluppi et al. provide resources and outline training needs and recommendations for junior researchers #GettingApplied nature.com/articles/s4146…
5.1K
Paul Scotti
@humanscotti
Aug 16, 2021
Now out in JEP:General! Visual working memory items drift apart (aka repulse/differentiate) due to active, not passive, maintenance. paulscotti.com/repulsion.pdf (half year late to share bc I was waiting for publication pdf, but realized I'm not allowed to share non-paywalled pdf🤷)
Paul Scotti
@humanscotti
Jan 19, 2021
1st project I started in my PhD, now in AP&P! rdcu.be/cdOa2 Summary: People memorized lots of colored real-world objects, unaware that many objects were same color. When tested on object colors, found swap & shift errors in relation to most commonly studied color. 1/
Paul Scotti
@humanscotti
Sep 25, 2023
BrainLM is a foundation fMRI model which can be fine-tuned for all sorts of specialized downstream tasks. We will have a public discussion on this paper this Fri 1pm ET on Zoom for @MedARC_AI NeuroAI journal club, feel free to join! How to join:
David van Dijk
@david_van_dijk
Sep 14, 2023
Introducing BrainLM 🧠🤖the first foundation model for #fMRI analysis trained on 6,700 hours of brain activity data! Fine-tune for specialized tasks or leverage zero-shot inference capabilities! @WuTsaiYale @YaleCompsci @YaleCBB @YaleMed biorxiv.org/content/10.110…
world-wide.org
Seminar Not Found
This seminar could not be located on World Wide.
16K