Search

NOMAD-EEG

NOMAD-EEG uses a diffusion backbone to build EEG foundations that are robust to electrode order and cap variations, processing signals with 3D coordinates and bipolar derivations, and leveraging state-space temporal modeling plus attention for cross-electrode integration; post-training, activations support a frozen probe or fine-tuning start.

[2611.08969] PhysioWave-T: Token-Efficient Wavelet Foundation Models for Biosignals

PhysioWave-T uses wavelet scale folding to reduce Transformer compute and memory for EEG, ECG, and sEMG models. Its EEG results lead on most reported comparisons, while reliability-aware fusion improves seizure detection on SeizeIT2.

[2610.35246] NOMAD-EEG: Native Omni-Montage Adaptive Diffusion

NOMAD-EEG is a small, diffusion-based EEG foundation model that is electrode-agnostic and patch-free. Trained on one corpus, it outperforms larger models across multiple EEG tasks and datasets, demonstrating strong seizure detection and epilepsy-related classification capabilities even with reduced electrode counts.

Chinese brain device converts thoughts into text with 94% accuracy – Tekniikan Maailma

Translated title. Original: Kiinalainen aivolaite kirjoittaa ajatuksista tekstiä 94 prosentin tarkkuudella – Tekniikan Maailma

Chinese researchers developed a scalp-based EEG electrode that maintained signal quality for hours. In tests, a participant used a virtual keyboard with over 94% accuracy, while standard gel electrodes degraded within four to five hours.

Deep learning model spots neonatal seizures from EEG brain maps with 97%

Automated neonatal seizure detection uses EEG-based brain topographic maps with deep transfer learning, achieving 97.1% subject-wise accuracy (8-second window) on data from 79 newborns; validated with unseen patients, the method aims to aid NICU monitoring and early intervention.

A new neurotechnology startup wants to read brain activity with ultrasonic headphones, without implants

Translated title. Original: Una nueva startup de neurotecnología quiere leer la actividad cerebral con unos auriculares de ultrasonidos, sin necesidad de implantes

Bridge Neurotech aims to build a non-invasive, portable brain-computer interface using functional ultrasound (fUSI) to observe brain activity from outside the skull, addressing limitations of implanted electrodes and external EEG. The approach relies on blood flow changes to infer neural activity, faces skull-related imaging challenges, and plans AI-based delay compensation; a early study in California is underway with healthy participants and implant recipients.

[2610.28083] IECG: Weakly-Supervised Interpretability-Enhancing Foundation Model for Electrocardiogram

A weakly-supervised ECG foundation model (IECG) jointly classifies diagnoses and localizes abnormalities without manual region annotations, using clinically guided attention and multimodal pretraining to provide interpretable waveform evidence while maintaining strong performance.

Brain–computer interface - Wikipedia

Invasive brain–computer interfaces face material and electronic challenges that affect long-term signal stability. Research targets reduce the body's reaction to implants and improve recording quality through flexible, tissue-like electrodes, better amplifier/ADC integration, and low-power designs; recent work demonstrates faster-than-before decoding of speech and motor intent using neural networks and related models.

NeuroTech in 5 Minutes: 7 Terms You Need So You Never Feel Lost Again

This guide explains seven terms for assessing brain-computer interface reports, including device maturity, electrode types, invasiveness, decoding metrics, and FDA pathways. Examples show why channel counts and accuracy figures need context, and why lab demonstrations are not the same as approved products.

Wearable sleep trackers are using your brainwaves to train AI models without explicit user consent — tens of millions of hours of EEG data harvested | Tom's Hardware

A Science investigation found that wearable EEG companies have collected tens of millions of hours of brain recordings, with some using the data to train AI models before clearly disclosing this practice. The report also describes privacy and security risks, emerging rules for neural data, and research applications for EEG models.