ARTFEED — Contemporary Art Intelligence

EEG Decoding of Grasp and Lift Parameters Using Attention-Based Model

ai-technology · 2026-07-30

A team of researchers has introduced three regression models—partial least squares regressor, multilayered perceptron, and attention-based regressor—to interpret various kinematic and kinetic movement parameters from EEG signals. They assessed the models using the WAY EEG GAL dataset under both subject-specific and subject-independent scenarios, employing two approaches: one model for all parameters and distinct models for each parameter. The attention-based regressor outperformed the others, achieving an R² of 0.8 and a latency of 29.2 milliseconds, marking a significant advancement in the simultaneous decoding of multiple parameters. This research tackles a crucial issue in brain-machine interfaces, particularly for aiding individuals with mobility challenges, including stroke survivors and amputees.

Key facts

  • Three regression models are proposed: partial least squares regressor, multilayered perceptron, and attention based regressor.
  • The models decode multiple kinematic and kinetic parameters from EEG signals.
  • Evaluation is performed on the WAY EEG GAL dataset.
  • Two strategies are tested: single model for all parameters and separate models for each.
  • The attention based regressor achieves an R² of 0.8 and a latency of 29.2 milliseconds.
  • The study focuses on subject-specific and subject-independent conditions.
  • The goal is to expand usability and control of brain-machine interfaces.
  • Target users include stroke survivors or amputees with limited mobility.

Entities

Sources