Alexis Thual

Our paper on using Optimal Transport to compare human cortical surfaces was accepted at #NeurIPS22 🥳 We implement a new OT solver (FUGW) to compare human brains although their anatomy and functional activity patterns differ a lot. arxiv.org/abs/2206.09398 github.com/alexisthual/fu…

@JoinStateChange @analog_kyle @rayanhindi Thanks a lot for having me Kyle, I had a wonderful time talking about what we do at karavela and watching the other episodes too!

Hosted by @analog_kyle , produced by @rayanhindi , and directed by Lancelot Bernheim , State Change Season 1 has taken us on an extraordinary journey into the outer edges of human possibility. From cryopreservation to rewriting our genes, from space exploration to AI doppelgängers, from mapping the mind to putting democracy under stress — this season explored what happens when science fiction starts becoming reality. Thank you to all our guests for making this first season possible. @emilkendziorra @CathyTie @santisiri @b_belvisi @arthaud_ @alexisthual @tomorrowbio @InterstellarLA And to all our teams in Paris, New York and Buenos Aires Stay tuned for season 2 #singularity #futuretech #humanity #ai #science

I'm proud we are releasing LAION-fMRI, a densely sampled 7T fMRI dataset of natural images, with very broad stimulus sampling for testing countless hypotheses & deeply exploring brain representations. It is now available at laion-fmri.hebartlab.com What does LAION-fMRI offer? 🧵

Face au péril imminent, mettre de la méthode "Éviter les triangulaires au 2nd tour en retirant le.la candidat.e moins bien placé.e et de soutenir partout le.la candidat.e des forces démocratiques restant face au RN" lemonde.fr/idees/article/…

@gabrielpeyre This may not directly solve your problem, but I needed to compute the sum over rows/cols of CSR matrices to implement the Sinkhorn scaling algorithm on sparse matrices. I think the following util function performs well: #L131" target="_blank" rel="nofollow noopener">github.com/alexisthual/fu…

I decided not to attend NeurIPS in person despite having two accepted papers. Why? The unsustainable impact of flying. Want to know more? Check out the 2-pager I wrote ⬇️ t.ly/Gc2RV --> I am open to presenting in Europe around conf week. Reach out if interested!

@russpoldrack I'm using google slides too, for lack of a pleasing open-source option. Thus far, the main pitfall was the inability to export PDF slides with videos (which is always tricky). I use GIFs instead and it works alright. Figma could be interesting maybe

All NeuroImage and NeuroImage:Reports editors have resigned over the high publication fee, and are starting a new non-profit journal imaging-neuroscience.org This comes with great regret, and a huge amount of thought and discussion- please read announcement to get more details.

You work in social sciences and need to collect data on Twitter before the ship sinks? Check out @medialab_ScPo #OpenSource tools like minet and collect away :) #python #webmining #datascience github.com/medialab/minet

Un cours vraiment incroyable pour toute personne francophone qui est intéressée de comprendre la recherche récente sur la façon dont le cerveau représente l’information ! On apprend comment sont représentés les visages, l’espace, et même un peu le langage.

New preprint!🚀🧠 “Information-Restricted Neural Language Models Reveal Different Brain Regions’ Sensitivity to Semantics, Syntax and Context” Using NLMs, we've developed a new method that reveals brain maps sensitivity to different aspects of language. arxiv.org/abs/2302.14389

We have 2 open calls for a fully-funded PhD in #NeuroAI at Ecole Normale Supérieure: 1. euraxess.ec.europa.eu/jobs/66405 w/ Pierre Bourdillon 2. euraxess.ec.europa.eu/jobs/67064 w/ Yair Lakretz.

A PhD position in Artificial Intelligence and Cognitive NeuroScience “Linking Linguistics and Brain Dynamics with Deep Language Models” at @ENS_ULM Co-supervised with @JeanRemiKing

@leonardhussenot @PradyuPrasad Here are a few more examples in which I tried to see what happens with gender, which is a simpler classification problem in French (sorry for non-French speakers). I'm not sure it's very informative, but it's interesting 😅

@leonardhussenot @PradyuPrasad I agree my experiment is not very fair😁 But I guess it takes more than prompting to assess whether there is information about letters in the embedding. I think training a classifier to detect words starting with a specific letter given activations of a given layer could help

@emrobSci Thank you for your support, and for you and your team's time and help when we were trying to tweak the hyper parameters of MSM!

Really excited to see this paper published - tackling the hugely difficult problem of learning functional correspondences been individuals with highly different cortical anatomies. Very much looking forward to reading this in detail!

Looking for an internship with: statistics, open-source dev, and interaction with medical topics and collaborators? With @JulieJosseStat we are looking for a research intern! The topic is about the (un)representativeness of clinical trials🔬📊 filesender.renater.fr/?s=download&to…

Interested in the anatomy and function of the prefrontal cortex across different species?

@AskDrJeg Moreover, I agree with you: evaluating gains on downstream tasks would be a big plus! We measured how our transport plans can align myelin maps (it's in the sup. mat.) and are working on evaluating gains on downstream classifiers / decoders😊

@AskDrJeg Assuming basic OT = optimizing for a Wasserstein (W) loss, you'd be aligning cortices based on functional data only. We know this wouldn't work since we can easily exhibit areas of the PFC and TC with similar functional data. We thus add a Gromov-W loss to preserve cortical geom

Our paper on using Optimal Transport to compare human cortical surfaces was accepted at #NeurIPS22 🥳 We implement a new OT solver (FUGW) to compare human brains although their anatomy and functional activity patterns differ a lot. arxiv.org/abs/2206.09398 github.com/alexisthual/fu…