Prewhitening high-dimensional FMRI data sets without eigendecomposition.

Area of Science:

• Data processing and analysis
• Computational neuroscience
• Machine learning

Background:

• Linear whitening is crucial for data preprocessing in various fields, including neuroimaging.
• Traditional methods often rely on truncated eigendecomposition (ED), which can be computationally intensive for high-dimensional data.
• Minimizing mean squared error is a key objective in data transformation tasks.

Purpose of the Study:

• To propose a computationally efficient linear whitening algorithm.
• To minimize the mean squared error between original and whitened data.
• To avoid the use of truncated eigendecomposition (ED).

Main Methods:

• The proposed algorithm utilizes Lanczos vectors to approximate dominant eigenvectors and eigenvalues of the covariance matrix.
• It focuses on achieving accurate whitening without full matrix decomposition.
• The method's performance is evaluated on a high-dimensional fMRI dataset.

Main Results:

• The algorithm accurately approximates major eigenvectors and eigenvalues.
• It achieves a significant reduction in computational cost compared to truncated ED.
• Experimental results on fMRI data demonstrate comparable accuracy to traditional methods.

Conclusions:

• The proposed linear whitening algorithm offers a computationally efficient alternative to truncated ED.
• It provides accurate data whitening for high-dimensional datasets.
• This method has potential applications in neuroimaging and other fields requiring efficient data preprocessing.