Adaptive regularized canonical correlations in clustering sensor data

A regularized canonical correlations scheme is proposed for adaptive clustering of sensor measurements according to their information content. A novel framework utilizing sparsity-inducing regularization and exponential weighing is designed to deal with nonstationary settings. Distributed recursions to minimize the proposed formulation are put forth by utilizing coordinate descent techniques combined with the alternating direction method of multipliers. Numerical tests demonstrate that the novel adaptive clustering framework is capable to deal with nonstationary settings while outperforming existing alternatives.