Multimodal approach to estimate the ocular movements during EEG recordings: A coupled tensor factorization method.

Area of Science:

• Data analysis
• Multivariate statistics
• Signal processing

Background:

• Coupled tensor factorization is a powerful technique for analyzing multi-modal data.
• Existing methods like advanced coupled matrix-tensor factorization (ACMTF) rely on strong assumptions.
• These assumptions are often not met in real-world datasets, limiting applicability.

Purpose of the Study:

• To propose a relaxed criterion for coupled tensor factorization (RACMTF).
• To enable more robust analysis of multi-modal data under weaker assumptions.
• To demonstrate the utility of RACMTF in scenarios with correlated factors and differing noise levels.

Main Methods:

• Development of the relaxed advanced coupled matrix-tensor factorization (RACMTF) criterion.
• Numerical simulations to validate the performance of RACMTF.
• Application of RACMTF to real-world Gaze&EEG data.

Main Results:

• The RACMTF criterion is based on weaker assumptions, making it more applicable to real data.
• Numerical simulations confirm the benefits of using joint datasets, particularly when factors are highly correlated.
• The method effectively handles situations where one data modality has less noise.
• Successful application to Gaze&EEG data for ocular artifact estimation.

Conclusions:

• The proposed RACMTF method offers a more flexible and robust approach to coupled tensor factorization.
• It enhances the analysis of multi-modal datasets, especially in the presence of correlated factors and noise.
• RACMTF provides a valuable tool for artifact detection and removal in neurophysiological recordings like EEG.