Fig. 2
From: Singular spectrum-based matrix completion for time series recovery and prediction
Example of the generative process of a real trajectory matrix from the Intel-Berkeley dataset. The matrix on the left is the Hankel matrix generated by a sensor for a given set of window and lag values. By utilizing the SSA-based dictionary, the mapping of the Hankel matrix results in an extremely low rank representation matrix shown in the middle, where a small number of singular values capture most of the signal energy, as shown in the right figure