eeg

Meta has improved Brain2QWERTY, a non-invasive system using MEG and EEG to decode brain activity into text, enabling typing.

This paper investigates whether EEG signals can complement eye-tracking signals for automatic keyphrase extraction from microblogs. Using the ZuCo corpus, the authors show that cognitive signals, especially EEG, improve AKE performance across different models.

The article presents NeuraDock Agent, an open-source architecture that integrates a deterministic EEG engine with an LLM interface, using hardware- and implementation-aware context to improve boundary awareness for low-channel EEG.

This tutorial paper presents NeuraDock Agent, an open-source EEG workflow for visual cognitive load analysis with alpha dynamics, including preprocessing, quality control, real-time API, and LLM interpretation.

NeuroSonic introduces a conditional flow-matching framework for reconstructing continuous speech from EEG signals, addressing the structural mismatch between neural and acoustic data by learning a deterministic probability-flow velocity field. It achieves up to 26.3% improvement in perceptual quality over existing GAN, diffusion, and mean-flow baselines on cross-subject benchmarks.

RECTOR is a self-supervised framework that learns joint region-channel-temporal representations from EEG/sEEG signals for affective and cognitive state classification, achieving state-of-the-art results on emotion recognition and task-engagement benchmarks.

Introduces Random Attention (RA), a lightweight temporal modeling module for mobile sleep staging that uses fixed random projections for similarity-based aggregation, achieving competitive performance with minimal additional parameters.

This paper investigates reducing the computational complexity of deep neural networks for EEG analysis on wearable devices by applying parameter quantization and electrode reduction techniques, demonstrating significant complexity reduction with minimal accuracy loss for epileptic seizure detection.

This study uses language model embeddings to quantify semantic association in self-paced reading and EEG data, examining how different implementations affect measures of reading difficulty.

This paper identifies and diagnoses the 'Identity Trap' in EEG foundation models, where high accuracy may stem from subject-identity features rather than genuine clinical biomarkers. It proposes FMScope, a frozen-representation protocol to disentangle these signals, and demonstrates that subject-identity confounding is universal across three models and removable with linear methods.

This paper presents a region-level evaluation framework for EEG-based cognitive workload prediction, showing that frontal electrode groups outperform full-scalp baselines by 15-20% in rank position while using fewer electrodes, supporting efficient workload monitoring systems.

This paper proposes a lightweight CNN architecture to improve adversarial robustness in EEG-based brain-computer interfaces, evaluating it against adversarial attacks and showing better classification performance than existing models.

This paper introduces Score-Guided Classification (SGC), a framework that models pathological priors using an unsupervised generative network for EEG-based depression detection, avoiding synthetic data augmentation and improving classification accuracy.

This paper discovers a shared valence axis (V-axis) across modern LLMs and human EEG signals, showing that a single direction from LLM internal representations aligns with neural responses to emotional stimuli. It also identifies the saturation regularity, explaining why LLM-derived supervision fails to improve EEG decoding and how leveraging residual diversity boosts performance.

A new dataset and model predict emotions from EEG data with over twice the performance of previous state-of-the-art.

This paper compares several post-hoc explainability methods applied to an InceptionTime model for EEG-based depression detection, finding partial convergence among methods while highlighting methodological variability and limitations.

This paper identifies and explains a spectral bias in reconstruction-based EEG foundation models, where embeddings over-represent aperiodic and low-frequency components while under-representing oscillatory components, especially at higher frequencies, leading to poor performance in low-resource settings.

This paper uses EEG recordings to study neural dynamics when humans process AI-generated hallucinated content, revealing distinct cognitive patterns and differences between misjudged and correctly judged hallucinations.

This paper shows that traditional machine learning models using spectral features from EEG signals can match or outperform state-of-the-art attention-based deep learning models for diagnosing neurodegenerative diseases, suggesting fundamental limitations of attention mechanisms in this domain.

This paper applies TopK Sparse Autoencoders to three EEG foundation models (SleepFM, REVE, LaBraM) to extract interpretable feature dictionaries and introduces a framework for concept steering, revealing representational failures and clinical entanglements.