Stéphane d'Ascoli
@stephanedascoli
Followers
1K
Following
138
Media
25
Statuses
113
Research Scientist @AIatMeta working on brain decoding. Prev @EPFL, @ENS_ULM, @NASA.
Paris, France
Joined November 2018
Think Transfomers are terrible at logical reasoning? Think again 💥 In this collaboration with Samy Bengio, @jsusskin (Apple) & Emmanuel Abbé (EPFL), we show that when trained with Boolean inputs and symbolic outputs, they become very powerful 🧠 https://t.co/QS24rcv79U 🧵⤵️
8
117
589
Very pleased to share our latest study!
Can AI help understand how the brain learns to see the world? Our latest study, led by @JRaugel from FAIR at @AIatMeta and @ENS_ULM, is now out! 📄 https://t.co/y2Y3GP3bI5 🧵 A thread:
8
40
526
Adapt your model, not your preprocessing! This week I’m in Strasbourg at #GRETSI2025 presenting our latest work with the Brain&AI team (@metaai) and the MIND team (@Inria).
1
3
12
🏆 We're thrilled to announce that Meta FAIR’s Brain & AI team won 1st place at the prestigious Algonauts 2025 brain modeling competition. Their 1B parameter model, TRIBE (Trimodal Brain Encoder), is the first deep neural network trained to predict brain responses to stimuli
149
425
3K
🙏Thanks to our wonderful team Jérémy Rapin, @BenchetritYoha1, @HubertJBanville, and @JeanRemiKing Thanks to the @AlgonautsPro organizers, and to @CNeuromod for making this possible, and the open source community on which this work is built!
0
0
1
📈Performance improves with additional data and context (>1000 words!) with no evidence of a plateau – indicating that future models of the human brain will improve as more data is used to train them. A promising result for the intersection of neuro and AI!
1
0
5
🧠We observe that this combination outperforms unimodal versions of the model in pretty much all brain areas (but V1), and especially in associative areas. This suggests a surprisingly distributed broadcast of all modalities across the cortex.
1
0
2
🧐The results show that each modality tends to peak in its expected brain region: visual representations in the visual cortex, audio representations in the auditory cortex, and language… pretty much everywhere else.
1
0
4
For this we built a TRIBE - Trimodal Brain Encoder, a 1B transformer which leverages three recent foundation models from @MetaAI: 🎥V-Jepa2: for videos 🔈Seamless’ Wav2vec2-Bert: for sounds 📚Llama 3.2: for text
1
0
3
⚽️The goal of this competition is to predict fMRI brain responses of 4 subjects who watched 80 hours of movies in the scanner – by far the largest longitudinal data of its kind.
1
1
1
🏆 Thrilled to announce we reached 1st position in the Algonauts 2025 Competition with our 1B model of the brain watching movies! 📄Paper: https://t.co/TxVNNzQyeF 🧑💻Code: https://t.co/m7km1HkbAJ 💿Data: https://t.co/Hsq9u0vIZG ⚔️Challenge: https://t.co/8ms3AOQNMC 👇Thread:
3
24
116
🔴 Our `Brain and AI` team at @MetaAI has a new Research Engineer position in Paris to work on infrastructure and scaling deep learning experimentations: https://t.co/8TO7nCPq3j We'll favor industry experience or Github track record.
0
4
33
Research paper from Meta FAIR and @bcbl_ researchers – Brain-to-Text Decoding: A Non-invasive Approach via Typing ➡️ https://t.co/6FXpmNiEgJ
122
354
2K
Congrats @JarodLevy and the rest of the team! What a start for your PhD 🤩
🔥”Brain-to-Text Decoding” is now out on ArXiv: https://t.co/xroRXp9MJm Our paper from @AIatMeta and @bcbl_ presents Brain2Qwerty, an AI model that decodes text from non-invasive recordings of the brain. Below a detailed thread: 🧵1/7
0
0
4
⚡️Check out our workshop tomorrow at @lpiparis_, with great speakers (@gabrielpeyre, @stephanedascoli, @TonyBonnaire + many more) covering Theory and Applications of Generative AI as well as Connexions with neuroscience 🧠 And there's food 🍰 ➡️ https://t.co/V2sTqZC4bl
0
4
7
Our work on predicting cell locations from gene expressions is finally out 🧬 Huge congrats to @mariabrbic and the rest of the team 🎉
Thrilled to share LUNA🌕 – our new generative AI model that reassembles tissues from dissociated cells! Think of it as AlphaFold for cells 🧬 From the MERFISH mouse brain atlas to the scRNA-seq atlas & Slide-tags datasets, LUNA is widely applicable. Bringing us closer to virtual
0
0
5
Our paper from @AIatMeta and @bcbl_ is out on arxiv 🔥 “From Thought to Action: How a Hierarchy of Neural Dynamics Supports Language Production” https://t.co/ZhDUr3ZGxS How does the brain transform a thought into a sequence of motor actions? Results summarized in 🧵1/8
1
1
6
Two new studies from our team we're particularly happy about Study 1: Brain-to-Text Decoding: https://t.co/KT9egURLRl Study 2: From Thought to Action: https://t.co/NBUuBb6Unm Blog: https://t.co/UnwK8YO6dt
5
87
290
Great work led by @LStoffl!
I'm thrilled to present our paper "Elucidating the Hierarchical Nature of Behavior with Masked Autoencoders" at #ECCV24! 🎉 We focus on self-supervised learning for action segmentation to uncover the hierarchical structure of behavior. Paper: https://t.co/N4UBgbpqwY 👇🧵
0
0
3
I'm thrilled to present our paper "Elucidating the Hierarchical Nature of Behavior with Masked Autoencoders" at #ECCV24! 🎉 We focus on self-supervised learning for action segmentation to uncover the hierarchical structure of behavior. Paper: https://t.co/N4UBgbpqwY 👇🧵
1
7
24
Missed our Oral and Poster at ICML 2024? Rewind the Arrow of Time, and take a look at https://t.co/R4WeEhJV3i ! @HonglerClement @jchwenger
0
3
10