BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//CERN//INDICO//EN
BEGIN:VEVENT
SUMMARY:Cross-Dataset EEG Workload Decoding Using Riemannian Deep Learning
  and Self-Supervised Representation Learning
DTSTART;VALUE=DATE-TIME:20260609T193000Z
DTEND;VALUE=DATE-TIME:20260609T195000Z
DTSTAMP;VALUE=DATE-TIME:20260606T094350Z
UID:indico-contribution-10361@events.saip.org.za
DESCRIPTION:Speakers: Chitaranjan Mahapatra (Institute for Basic Science (
 IBS)\,Daejeon 34126\, South Korea)\nReliable decoding of cognitive workloa
 d from electroencephalography (EEG) signals is essential for adaptive robo
 tics\, neuroergonomics\, intelligent transportation\, and human–machine 
 interaction systems. Despite recent advances in deep learning\, EEG-based 
 workload classification remains limited by poor cross-subject and cross-da
 taset generalization. In this work\, we investigate hybrid Riemannian deep
  learning and self-supervised representation learning frameworks for robus
 t EEG workload estimation across heterogeneous datasets. We developed a co
 mprehensive benchmark pipeline integrating classical signal processing\, c
 ovariance-based Riemannian geometry\, transformer models\, self-supervised
  contrastive learning\, and explainable artificial intelligence. Experimen
 ts were performed using the publicly available ds007262 arithmetic EEG wor
 kload dataset and the STEW simultaneous task EEG workload dataset. Six dec
 oding frameworks were evaluated: FBCSP\, EEGNet\, Riemannian tangent-space
  classifiers\, transformer networks\, self-supervised learning architectur
 es\, and a proposed hybrid Riemannian deep learning model. Subject-indepen
 dent evaluation using leave-one-subject-out cross-validation demonstrated 
 that hybrid covariance-aware models consistently outperformed conventional
  CNN pipelines. The proposed hybrid framework achieved approximately 61% m
 ean classification accuracy\, while transformer and self-supervised learni
 ng models demonstrated improved temporal representation learning capabilit
 ies. Statistical benchmarking using ROC curves\, precision–recall curves
 \, violin plots\, and Wilcoxon significance testing confirmed the superior
 ity of hybrid geometric learning approaches. Explainable AI analyses revea
 led neurophysiologically meaningful workload-related EEG patterns involvin
 g frontal theta enhancement and parietal alpha suppression. Channel import
 ance mapping demonstrated dominant contributions from frontal and parietal
  cortical regions associated with attentional control and working memory p
 rocesses. To evaluate robustness and transferability\, cross-dataset exper
 iments were performed by training on ds007262 and testing on STEW. Results
  demonstrated substantial performance degradation caused by domain shift\,
  recording variability\, and montage mismatch across datasets. These findi
 ngs highlight the major challenge of EEG transferability in real-world app
 lications and emphasize the need for future domain adaptation and transfer
  learning strategies. The proposed benchmark framework provides a reproduc
 ible pipeline for evaluating EEG workload decoding algorithms under both s
 ubject-independent and cross-dataset settings. The integration of Riemanni
 an geometry\, explainable AI\, and self-supervised learning offers a promi
 sing direction for robust cognitive monitoring systems in intelligent robo
 tics and adaptive neurotechnology applications.\n\nhttps://events.saip.org
 .za/event/274/contributions/10361/
LOCATION:
URL:https://events.saip.org.za/event/274/contributions/10361/
END:VEVENT
END:VCALENDAR