Advancing SSVEP-BCI Decoding: Cross-Subject Transfer Learning and Short Calibrated Approach with ELM-AE

The Steady-State Visually Evoked Potential (SSVEP) is a robust paradigm for developing a high-speed Brain-Computer Interface (BCI). However, one of the challenges of BCI is to face the variability of EEG signals between subjects to reduce or eliminate the time calibration process for a new subject (...

Full description

Saved in:
Bibliographic Details
Published in:2024 46th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC) Vol. 2024; pp. 1 - 5
Main Authors: Flores, Christian, Casas, Paolo, de Carvalho, Sarah Negreiros, Attux, Romis
Format: Conference Proceeding Journal Article
Language:English
Published: United States IEEE 01.07.2024
Subjects:
Accuracy
Adult
Algorithms
Brain-Computer Interface
Brain-Computer Interfaces
Calibration
Electroencephalography
Electroencephalography - methods
ELM-AE
Engineering in medicine and biology
Evoked Potentials, Visual - physiology
Extreme learning machines
Face recognition
Humans
Machine Learning
Male
Nonlinear Transformation
Signal Processing, Computer-Assisted
SSVEP
Steady-state
Target recognition
Transfer learning
ISSN:2694-0604, 2694-0604
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The Steady-State Visually Evoked Potential (SSVEP) is a robust paradigm for developing a high-speed Brain-Computer Interface (BCI). However, one of the challenges of BCI is to face the variability of EEG signals between subjects to reduce or eliminate the time calibration process for a new subject (target subject). Some approaches propose linearly transforming; however, it limits the ability to capture complex and nonlinear relationships in data. This study presents a method for performing a Nonlinear Transformation (NLT) using an Extreme Learning Machine Autoencoder (ELM-AE) on SSVEP trials. To improve the NLT, it maps each trial from the existing subjects (source subjects) to one or a few templates from the target subject. This approach can enhance cross-subject recognition classification, reducing the calibration time for the target subject. Our results reported that, for one template, NLT and LST achieved 84.23% and 82.19% average recognition accuracy, respectively. Thus, our results reported that the recognition accuracy of NLT outperformed LST for all template sizes across all 35 subjects. These results demonstrated the feasibility of the NLT using one or a few templates for rapid calibration for the target subject.
Bibliography:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 23
ISSN:2694-0604
2694-0604
DOI:10.1109/EMBC53108.2024.10782739