Multiview Bayesian Correlated Component Analysis

Area of Science:

• Neuroscience
• Computational Neuroscience
• Statistical Analysis

Background:

• Correlated Component Analysis (CCA) investigates brain process similarity using multiple stimulus views.
• CCA assumes identical spatial networks for component identification.
• Existing methods like Canonical Correlation Analysis (CCA) and Correlated Component Analysis (CCA) offer limited flexibility in assessing representation universality.

Purpose of the Study:

• To propose a hierarchical probabilistic model for inferring the level of universality in multiview brain data.
• To introduce Bayesian Correlated Component Analysis (BCCA) as a flexible tool for analyzing neural representations.
• To evaluate BCCA's performance against existing algorithms and validate it on real-world EEG data.

Main Methods:

• Developed a hierarchical probabilistic model, Bayesian Correlated Component Analysis (BCCA).
• Simulated multiview data to compare BCCA with three other relevant algorithms.
• Utilized a benchmark Electroencephalography (EEG) dataset for further validation.

Main Results:

• BCCA successfully infers the level of universality in neural representations, ranging from unrelated to identical.
• BCCA demonstrated superior performance compared to three other algorithms in simulated data.
• Analysis of EEG data revealed variability in spatial representations across multiple subjects.

Conclusions:

• Bayesian Correlated Component Analysis (BCCA) provides a robust framework for assessing the universality of neural representations.
• The model offers a more nuanced understanding of brain process similarity than traditional methods.
• BCCA is a valuable tool for analyzing complex multiview neuroimaging data and understanding inter-subject variability.