romquentin.bsky.social

Very proud of Coumarane Tirou on his first first-author paper, now out in Nature Communications. On how the brain learns hidden regularities through noise using two distinct neural mechanisms. Huge congrats !! 🥳

Coumarane Tirou@coumt.bsky.social

🚨 New paper alert! Happy to share my first paper, "Learning regularities in noise engages both neural predictive activity and representational changes", has been published in Nature Communications. www.nature.com/articles/s41...

🧠 Does the brain pre-activate feature-specific patterns resembling the perception of an upcoming, predictable stimulus? We reanalyzed the open dataset from Demarchi et al. (2019) and found no evidence of feature-specific neural pre-activation during auditory predictions. 🧵

github.comGitHub - MEL-Eduwell-lab/predictive_activityContribute to MEL-Eduwell-lab/predictive_activity development by creating an account on GitHub.

We also disagree with the core claim that feature-specific pre-activation exists in the original dataset. This debate only exists because they shared their data openly. That openness is worth celebrating! 🎉 Code: github.com/MEL-Eduwell-...

nature.comNo evidence of neural feature-specific pre-activation during the prediction of an upcoming stimulus - Nature CommunicationsNature Communications - No evidence of neural feature-specific pre-activation during the prediction of an upcoming stimulus

Peer-review files are available for the full debate (www.nature.com/articles/s41...)

I.e. whether the brain pre-activates patterns similar to those evoked while actually perceiving the same stimulus. For this specific claim, there is no evidence in the dataset. This does not, of course, challenge the broader and well-established literature on brain predictions in general.

We disagree with the response by Demarchi et al., including unsupported claims: alleged selective interpretation, and alleged data leakage (there is none — check our code).

doi.orgNo evidence of neural feature-specific pre-activation during the prediction of an upcoming stimulus - Nature CommunicationsNature Communications - No evidence of neural feature-specific pre-activation during the prediction of an upcoming stimulus

Full paper: doi.org/10.1038/s414... Authors: @oabdoun.bsky.social, Dmitrii Todorov, Arnaud Poublan-Couzardot, @coumt.bsky.social, @antoinelutz.bsky.social, Marine Vernet, @romquentin.bsky.social

In their response, Demarchi et al. propose alternative analyses to identify neural predictions. To be clear: our reanalysis specifically concerns predictions that resemble perception, as in the original 2019 paper.

The original result is fully explained by a statistical artifact: the interaction between transition probabilities in structured sequences and classifier confusion bias.

We show this two ways: → Empirically, by reordering random sequences to mimic structured ones → Theoretically, with an analytic expression for the true chance level

github.comGitHub - MEL-Eduwell-lab/DeepEpiXContribute to MEL-Eduwell-lab/DeepEpiX development by creating an account on GitHub.

Introducing DeepEpiX ! An open-source software for automated epileptic spike detection in MEG recordings ( soon EEG)! Visualization + annotation + 3 deep learning models, all in one GUI. 📄 DeepEpiX paper: doi.org/10.1016/j.jn... 💻 Software & code: github.com/MEL-Eduwell-...

Detecting interictal spikes (IEDs) in MEG is a important step in the presurgical evaluation of drug-resistant epilepsy, but it's time-consuming and subject to inter-rater variability. DeepEpiX is built on MNE-Python, designed to be modular and extensible, and freely available on GitHub.

With Agnès Guinard, Julien Jung, Romain Bouet, Elias Boulanger, Julien Coudray, Denis Schwartz, Sébastien Daligault, Aurore Semeux-Bernier, Christian-George Bénar, Francesca Bonini, Romain Quentin and Pauline Mouchès #Epilepsy #MEG #DeepLearning #Neuroscience #OpenSource #ClinicalNeuroscience

github.comGitHub - MEL-Eduwell-lab/DeepEpiXContribute to MEL-Eduwell-lab/DeepEpiX development by creating an account on GitHub.