Journal: Frontiers in Neuroscience

Article Title: Memory Efficient PCA Methods for Large Group ICA

doi: 10.3389/fnins.2016.00017

Figure Lengend Snippet: Estimation error as compared to EVD (IRAM) . L 2 -norm of error is computed between the eigenvalues of each method and the eigenvalues of the EVD method.

Article Snippet: With less than 10,000 time points, the first PCA step could be easily solved by loading the data in blocks along the voxel dimension, summing covariance matrices of dimension t × t across blocks, i.e., ∑ n = 1 b l o c k s ( C t t ) n = F Λ F T , and using the EVD (IRAM) method [ eigs (·) function in MATLAB].

Techniques:

Journal: Frontiers in Neuroscience

Article Title: Memory Efficient PCA Methods for Large Group ICA

doi: 10.3389/fnins.2016.00017

Figure Lengend Snippet: Computing time (in minutes) taken to solve group-level PCA using EVD (IRAM), Large PCA, MPOWIT, SVP, and STP algorithms . Using different numbers of subjects and components. The computing time of both Large PCA (un-stacked) and MPOWIT (un-stacked) are also reported.

Article Snippet: With less than 10,000 time points, the first PCA step could be easily solved by loading the data in blocks along the voxel dimension, summing covariance matrices of dimension t × t across blocks, i.e., ∑ n = 1 b l o c k s ( C t t ) n = F Λ F T , and using the EVD (IRAM) method [ eigs (·) function in MATLAB].

Techniques: