Regularized canonical correlations for sensor data clustering

The task of determining informative sensors and clustering the sensor measurements according to their information content is considered. To this end, the standard canonical correlation analysis (CCA) framework is equipped with norm-one and norm-two regularization terms to estimate the unknown number of field sources and identify informative groups of sensors. Coordinate descent techniques are combined with the alternating direction method of multipliers to derive an algorithm that minimizes the regularized CCA framework. An efficient scheme to properly select the regularization coefficients associated with the norm-one and norm-two terms is also developed. Numerical tests corroborate that the novel scheme outperforms existing alternatives.