Sparse canonical correlation analysis algorithm with alternating direction method of multipliers

摘要

Sparse canonical correlation analysis (SCCA) is applied to estimate the optimal combination coefficients with sparse properties between two multivariate datasets. It reduces the probability of some unimportant variables and can be interpreted more easily than non-sparse canonical correlation analysis (CCA), especially for high-dimensional datasets. In this paper, a new SCCA algorithm, based on the structure of alternating direction method of multipliers (ADMM), is proposed to solve thel(1)-penalized SCCA optimization problems. Then, a local subproblem is solved effectively by the variable-separation strategy, with explicit expressions, to update combination coefficients. Finally, several experiments validate that the proposed method is superior than the state-of-the-art SCCA algorithms.