Self-attentive Bidirectional LSTM Networks for Temporal Decoding of EEG Motor States

Authors: Sara Kamali • Fabiano Baroni • Pablo Varona

Using any code or material in this repository requires prior permission from the authors.
If you reference the methods or results, please cite the paper as listed above.
MATLAB and Python source files are provided.

---

1. Dataset

• Task: stereotypical finger-pinching
• Public dataset DOI: https://doi.org/10.1093/gigascience/gix034
• Context: Fig. 1 of the paper shows the task timeline and execution steps.

---

2. Pre-processing (MATLAB)

File	Purpose
MS_cls_info.mat	Subject / component indices
initialize_environment.m	Parameter settings
preparing_data_format_and_labels.m	Re-formats data and labels for Python

Detailed EEG pre-processing procedures are available in the companion repository:
https://github.com/GNB-UAM/Kamali_Mu_and_beta_power_precue_effects_double_dissociate_latency

---

3. Deep-learning model (Python)

• Script: BiLSTM_Self_attention_EEG_classifier.py
  • Implements a BiLSTM followed by a self-attention layer
  • Generates all evaluation metrics and plots reported in the paper

---

4. Additional material

File	Description
IWANN2025_Presentation_Kamali_etal.pdf	Slide deck used for the IWANN 2025 conference presentation

---

5. Reference

Kamali S., Baroni F., & Varona P. (2025).
Self-attentive Bidirectional LSTM Networks for Temporal Decoding of EEG Motor States.