Antoine Collas

SKADA-Bench : Benchmarking Unsupervised Domain Adaptation Methods with Realistic Validation On Diverse Modalities, has been published published in TMLR today 🚀. It was a huge team effort to design (and publish) an open source fully reproducible DA benchmark 🧵1/n. openreview.net/forum?id=k9F...

Physics-informed and Unsupervised Riemannian Domain Adaptation for Machine Learning on Heterogeneous EEG Datasets Combining electroencephalogram (EEG) datasets for supervised machine learning (ML) is challenging due to session, subject, and device variability. ML algorithms typically require identical features at...

Our EEG montage interpolation method is now in MNE-Python 1.10!

Based on our EUSIPCO 2024 paper, .interpolate_to() maps signals across caps in one line—ideal for preprocessing EEG before training AI models across datasets.

📄 arxiv.org/abs/2403.15415
🧠 mne.tools/stable/auto_...

#EEG #MNEPython #AI

Skada Sprint Alert: Contribute to Domain Adaptation in Python

📖 Machine learning models often fail when the data distribution changes between training and testing. That’s where Domain Adaptation comes in — helping models stay reliable across domains.

Opinion of the day: we don't desk reject enough in ML. Too much energy is wasted in 4x reviewing papers that will *obviously* be rejected.

Second opinion otd: we don't teach enough students to be positive. We should not seek how to reject a paper, but how to accept it.

And yes, #1 has a role in #2

We have been reworking the Quickstart guide of POT to show multiple examples of OT with the unified API that facilitates access to OT value/plan/potentials. It allows to select regularization/unbalancedness/lowrank/Gaussian OT with just a few parameters. pythonot.github.io/master/auto_...

Screengrab of the SVM-KM page

It's been 20 years and I think the new generation need to know about the SVM-KM toolbox. It was a Matlab open source SVM toolbox created in 2005 by @scanu.bsky.social, Yves Grandvalet, Vincent Guige, and Alain Rakotomamonjy 1/n github.com/rflamary/SVM...

@tommoral.bsky.social @agramfort.bsky.social

We put lots of effort to benchmark domain adaptation on many modalities👇🏻👇🏻

Gaël Varoquaux, vedette de l’intelligence artificielle et défenseur du logiciel libre L’informaticien et chercheur à l’Inria est l’expert français le plus cité dans les publications scientifiques portant sur l’IA. Avec Scikit-learn, un programme de machine learning dont il est le cocré...

Merci @lemonde.fr pour un joli résumé de mes aventures scientifiques et logiciels 📈📠
www.lemonde.fr/sciences/art...

Beaucoup de messages qui me tiennent à cœur : travail d'équipe, logiciel libre, rigueur scientifique

Merci aux collègues et amis qui ont témoigné, je suis ému de lire

Super proud of this work! DA is the way to go for many applications and Skada will democratize it!

🚀 Skada v0.4.0 is out!

Skada is an open-source Python library built for domain adaptation (DA), helping machine learning models to adapt to distribution shifts.
Github: github.com/scikit-adapt...
Doc: scikit-adaptation.github.io
DOI: doi.org/10.5281/zeno...
Installation: `pip install skada`

7/ GOPSA offers a robust solution for addressing joint (X, y) shifts in EEG data, significantly improving model generalization. It integrates Riemannian geometry with mixed-effects modeling, providing novel tools for generating neuroscientific and clinical insights.

6/ 🎯 Re-centering sites helped reduce the shift in X, while not placing all sites at the exact same reference point helped manage the shift in y. This preserves the statistical associations between X and y, improving model generalization.

5/ 📊 Our benchmark on the HarMNqEEG dataset showed significant improvement in performance for three metrics across most site combinations of GOPSA compared to baseline methods.

4/ 📊 Results on simulated data show how GOPSA adapts to various degrees of shifts in X and y. Panel C illustrates that GOPSA is specifically designed to handle joint shifts in X and y.

3/💡Our method, GOPSA, jointly learns parallel transport along a geodesic for each domain and a global regression model common to all domains, assuming that the mean y can be estimated.

2/ Multicenter neuroscience datasets often face shifts in both feature (X) and label (y) distributions. Our work specifically addresses cases where features are covariance matrices and focuses on joint (X, y) dataset shifts in EEG data.


Geodesic Optimization for Predictive Shift Adaptation on EEG data Electroencephalography (EEG) data is often collected from diverse contexts involving different populations and EEG devices. This variability can induce distribution shifts in the data $X$ and in the b...

Next week, we’ll present our spotlight paper at #NeurIPS2024 on domain adaptation for EEG data. Join us in East Exhibit Hall A-C on Friday at 4:30 PM!

arxiv.org/abs/2407.03878

Apolline Mellot @sylvchev.bsky.social @agramfort.bsky.social @dngman.bsky.social

A thread: 1/7

Good, published, benchmarks of machine learning / data science is crucial.

But so hard.
Well-cited "SOTA" methods typically crash often. They tend to be very computational expensive. Both make a systematic study impossible.

Finally, reviewers always ask for more methods, and more "SOTA".

@rflamary.bsky.social @ambroiseodt.bsky.social shared feelings 😅

Apolline Mellot, Antoine Collas, Sylvain Chevallier, Alexandre Gramfort, Denis A. Engemann
Geodesic Optimization for Predictive Shift Adaptation on EEG data
https://arxiv.org/abs/2407.03878