AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

EEG-X: Toward Device-Agnostic and Noise-Robust Foundation Models for EEG

Navid Mohammadi Foumani, Soheila Ghane, Nam Nguyen, Mahsa Salehi, Geoffrey Webb, Geoffrey Mackellar

NeurIPS 2026 · Label-efficient and robust learning › Recognition and long-tail · approach: Dense features and foundation priors

Read the paperCheck your idea against 137,217 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes EEG-X, a device-agnostic and noise-robust foundation model for EEG aimed at handling electrode/device variability and low signal-to-noise ratio.

Problem

EEG foundation models are immature due to cross-dataset variability from differing recording devices and electrode configurations, and the low SNR of EEG where neural activity is buried under artifacts and non-brain sources.

Approach

A foundation model for EEG designed to be device-agnostic (robust to electrode configurations) and noise-robust; the abstract is truncated so the specific mechanism is not stated.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add location-based channel embedding instead of none (ablation)EEG-X · cross-domain average (Crowdsourced, BCICIV_2a, SSVEP)accuracy: 40.02 -> 57.17 on SSVEP; average drop 8.81 without itImproved
use location-based instead of learnable channel embedding (ablation)EEG-X · cross-domain SSVEPbalanced accuracy: 41.2 -> 57.17Improved
use location-based instead of learnable channel embedding (ablation)EEG-X · in-domain averageaccuracy: drop of 0.43 with learnable CE vs EEG-X average 77.66Improved
use location-based instead of learnable channel embedding (ablation)EEG-X · TUEV in-domainaccuracy: 90.46 (no CE) / 90.14 (learnable CE) -> 91.15Mixed
add DiCT layer before MSE reconstruction loss (ablation)EEG-X · in-domain STEWaccuracy: 76.91 -> 80.32Improved
add DiCT layer before MSE reconstruction loss (ablation)EEG-X · cross-domain SSVEPbalanced accuracy: 51.55 -> 57.17Improved
reconstruct ICA-cleaned signal instead of raw signal (ablation)EEG-X · in-domain STEWaccuracy: 77.90 -> 80.32Improved
reconstruct ICA-cleaned signal instead of raw signal (ablation)EEG-X · cross-domain BCICIV_2abalanced accuracy: 35.31 -> 37.58Improved
add raw-space reconstruction alongside latent-space reconstruction (ablation)EEG-X · in-domain STEWaccuracy: 75.92 -> 80.32Improved
use ICA as preprocessing instead of raw inputEEG-X · STEWaccuracy: 75.12 -> 70.68Hurt
use ICA as preprocessing instead of raw inputEEG-X · TUEVaccuracy: 88.24 -> 83.36Hurt
use ICA as preprocessing instead of raw inputEEG-X · DREAMER and TUABaccuracy: 55.69 -> 56.24 (DREAMER); 78.12 -> 78.25 (TUAB)No clear effect

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy