Reconstructing human gaze behavior in from EEG in Stroop test using inverse reinforcement learning

Decoding Human Gaze Behavior in Stroop Tasks Using Inverse Reinforcement Learning from EEG Data Understanding human gaze behavior during cognitively demanding tasks offers valuable insights into underlying cognitive processes and potential clinical applications. This dissertation presents a novel framework for reconstructing human gaze behavior in the Stroop test by integrating electroencephalography (EEG) data with inverse reinforcement learning (IRL). We employed Generative Adversarial Imitation Learning (GAIL) combined with Proximal Policy Optimization (PPO) to model gaze trajectories, using EEG signals to guide the learning process. Data were collected from ten adult participants performing the Stroop test while their eye movements and EEG signals were recorded. The EEG data were preprocessed to extract key features indicative of cognitive load and attentional shifts. Eye-tracking data provided fixation locations and durations, which were mapped onto a grid-based representation of the visual field. Our results demonstrate that the EEG-informed IRL framework successfully reconstructed gaze patterns that closely resemble human scanpaths during the Stroop task. Metrics such as Target Fixation Probability and MultiMatch analysis confirmed the alignment of the predicted gaze paths with actual human behavior. The integration of EEG data enhanced the model's ability to adapt to cognitive load variations, reflecting the dynamic nature of human attention. This study contributes a methodological advancement in cognitive neuroscience by combining EEG and eye-tracking data within an IRL framework to decode and reconstruct gaze behavior. The findings have significant implications for understanding attentional mechanisms and cognitive control and offer potential applications in diagnosing and monitoring neurodegenerative disorders affecting executive function.