Título: Evaluating Multitask Learning for EEG Denoising and Classification Using Synthetic Data
Autores: Renato Botter Maio Lopes Rodrigues, Denis Gustavo Fantinato
Resumo: When developing Electroencephalography (EEG)-based Brain-Computer Interface (BCI) systems, it is important to identify meaningful attributes from the EEG signals that could be used to train reliable classifiers and accurately decode brain patterns to control external devices. Multitask Learning (MTL) is a machine learning technique that can be employed to address this challenge, by simultaneously performing two or more tasks, such as denoising EEG signals and learning latent representations in order to train a classifier to predict different brain states. In light of this, in the present work we have created a Machine Learning (ML) model inspired on EEGNet capable of denoising and classifying EEG signals and a framework to generate synthetic data for evaluation. Results suggest that the MTL model was able to properly denoise and classify the EEG signals. The similar classification performance obtained using Single-Task Learning (STL) on a noisy version of the signal, when compared with the MTL model used to classify and denoise EEG signals suggest that the temporal and spatial convolutional filtering blocks may be effectively mitigating the impact of the noise.
Palavras-chave: EEG; Brain-Computer Interface; Motor Imagery; Deep Learning.
Páginas: 7
Código DOI: 10.21528/CBIC2025-1156806
Artigo em PDF: CBIC_2025_paper1156806.pdf
Arquivo BibTeX:
CBIC_2025_1156806.bib