A research team in Brest, France

BROADERARTIFICIALINTELLIGENCE.

We study representations that help intelligent systems learn, adapt, and make sense of complex data across signals, images, audio, and brain activity.

Selected work All publications

EEG & Neuroimaging Few-Shot & Vision Audio & Signal Edge & Compression

BRoader Artificial INtelligenceIMT Atlantique / Lab-STICCScroll to explore ↓

01 / RESEARCH

Research

Bridging deep learning, signal processing, and mathematical representations.

6 Permanent Faculty

15+ PhDs & Postdocs

148 HAL Publications

3 Research Pillars

BRAIN is a research team of Lab-STICC (CNRS UMR 6285) hosted at IMT Atlantique and part of the Mathematical and Electrical Engineering (MEE) department.

The purpose of BRAIN is to investigate key questions at the crossbreed of Artificial Intelligence, Deep Learning and Signal Processing, with applications using images, sounds, text and more complex domains including neuroimaging data.

01Few-Shot ↗

Thrifty AI

How to find efficient representations in the low data regime (typically few-shot or few-label learning).

02Edge AI ↗

Compression of AI

How to find efficient representations to lower the demand in computations and energy of AI systems, typically for AI on edge.

03Manifolds ↗

Geometry and AI

How to leverage the geometry of latent spaces or input domains to better adapt and monitor AI systems.

02 / SELECTED WORK

Selected publications

A selection of recent highlights from the team. Filter by area or browse our full HAL archive.

All 148 publications

NEURIPS 202501 / EEG FOUNDATION MODEL

REVE: A Foundation Model for EEG

A 4D spatial-temporal foundation model pretrained on 60,000+ hours of EEG designed to adapt seamlessly to arbitrary electrode setups and downstream BCI tasks.

Yassine El Ouahidi, Jonathan Lys, Philipp Thölke, Nicolas Farrugia, Bastien Pasdeloup, Vincent Gripon, Karim Jerbi, Giulia Lioi

Paper ↗Project ↗

CVPR 202502 / VISION & FEW-SHOT

ProKeR: A Kernel Perspective on Few-Shot Adaptation of Large Vision-Language Models

Reformulating few-shot adaptation of CLIP-like models through proximal kernel ridge regression in feature space for fast, training-free generalization.

Yassir Bendou, Amine Ouasfi, Vincent Gripon, Adnane Boukhayma

Paper ↗Code ↗

ICASSP 202603 / PARAMETER-EFFICIENT AI

TensLoRA: Tensor Alternatives for Low-Rank Adaptation

Leveraging higher-order tensor factorizations across attention projections and layers to achieve superior compression-performance trade-offs when fine-tuning transformers.

Axel Marmoret, Réda Bensaid, Jonathan Lys, Vincent Gripon, François Leduc-Primeau

PDF ↗arXiv ↗

TMLR 202604 / ARCHITECTURE SEARCH