EEG

Lightning talk on LuMamba — using LeJEPA “world-model” self-supervision to pretrain a tiny, montage-agnostic foundation model for biosignals.

First evaluation of EEG foundation models for burst-suppression detection in ICU monitoring.

We bring LeJEPA to biosignals for the first time and show that the choice of self-supervised objective (masked reconstruction, LeJEPA, or a mix) trades latent-space structure for cross-montage generalisation, all on a 4.6M-parameter bi-Mamba model that is 377× cheaper than LaBraM.

How learned queries and cross-attention let LUNA decouple compute from electrode count, cutting FLOPs by 300× and GPU memory by 10× while still hitting state-of-the-art on TUAR.

5.4M-parameter multimodal biosignal foundation model (EEG/ECG/PPG) deployable on a RISC-V microcontroller.

Topology-invariant, linear-complexity EEG foundation model; first systematic study of LeJEPA for biosignals.

Physiologically-aware pre-training and INT8 deployment of a bidirectional Mamba EEG foundation model on a microcontroller.

Poster presentation of LUNA—our topology-agnostic EEG foundation model—at NeurIPS 2025 in San Diego.

Topology-agnostic EEG foundation model that delivers state-of-the-art performance with 300× fewer FLOPs and 10× lower memory usage.

Bidirectional Mamba EEG foundation model that matches transformer accuracy while scaling linearly with sequence length.