Predicting fixations and gaze location from EEG.

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Eye-tracking is crucial for understanding cognitive processes.
• Traditional eye-tracking requires specialized equipment.
• Estimating eye-gaze from electroencephalography (EEG) offers a non-invasive alternative.

Purpose of the Study:

• To investigate the feasibility of predicting eye-gaze and estimating fixation using deep learning models with EEG data.
• To explore the impact of spatial and temporal data dimensions, local vs. global processing, and architectural design on model performance.
• To compare proposed Transformer and LSTM-based architectures against state-of-the-art methods under reduced data constraints.

Main Methods:

• Development and comparison of two deep learning architectures: Transformer-based and LSTM-based.
• Evaluation of models on fixation prediction and gaze estimation tasks using EEG signals.
• Testing model robustness under reduced EEG signal length and channel count.

Main Results:

• The Transformer-based model outperformed the LSTM-only model.
• Both models achieved results comparable to or better than state-of-the-art approaches when trained from scratch.
• The Transformer model demonstrated higher sensitivity to signal length and channel reduction.

Conclusions:

• Deep learning models, particularly Transformers, can effectively predict eye-gaze from EEG data.
• Architectural design choices significantly impact performance, especially under data constraints.
• This research advances the potential for non-invasive, equipment-free eye-tracking solutions.