Conceitos Básicos
Exploring the potential of Joint-Embedding Predictive Architectures (JEPAs) in EEG signal encoding.
Resumo
This article delves into the use of Joint Embedding Predictive Architectures (JEPAs) for seamless cross-dataset transfer in EEG signal processing. It introduces Signal-JEPA, a novel approach for representing EEG recordings with domain-specific spatial block masking and downstream classification architectures. The study evaluates models on three different BCI paradigms: motor imagery, ERP, and SSVEP, highlighting the importance of spatial filtering for accurate classification. The research aims to bridge the gap in exploring block masking strategies over EEG channels to enhance dynamic spatial filtering. It investigates fine-tuning strategies and pre-trained SSL models' effectiveness across various BCI paradigms. The S-JEPA framework is detailed, emphasizing its training process and components like local encoder, contextual encoder, and predictor.
Estatísticas
"The study is conducted on a 54 subjects dataset."
"The downstream performance of the models is evaluated on three different BCI paradigms: motor imagery, ERP, and SSVEP."
Citações
"The potential of JEPA-like frameworks has been highlighted by their promising results with images."
"Applications to the EEG Domain of masking-based SSL techniques have started to emerge."