Sparse Canonical Correlation Analysis (SCCA) has been proposed to find pairs of sparse weight vectors that maximize correlations between sets of paired variables. This is done by computing one weight vector pair, deflating the correlation matrix between the views, and then repeating the process to compute the next pair. However, the deflation step used does not guarantee the orthogonality of the vector pairs. This is a very important requirement if one wishes to study the space spanned by these vectors, which should have very promising neuroscience applications. In the present work, we propose a new method for performing the deflation step in SCCA models. The ability of these vector pairs to generalize to new data was tested using an open-access dementia dataset which included T1-weighted MRI images and clinical information. The proposed method provided weight vector pairs that were both orthogonal and able to generalize to new data.

Link to Paper »


Jo~ao M Monteiro, Anil Rao, John Ashburner, John Shawe-Taylor, Janaina Mour~ao-Miranda




, 60–68.


Applied physics, Complex issues, Complex networks, Complex Systems, Computational Finance, and Computational Science