Hierarchical sparse dictionary learning (HiSDL) for heterogeneous high-dimensional time series
A system, method and computer program product for hierarchical sparse dictionary learning (“HiSDL”) to construct a learned dictionary regularized by an a priori over-complete dictionary, includes providing at least one a priori over-complete dictionary for regularization, performing sparse coding of the at least one a priori over-complete dictionary to provide a sparse coded dictionary, using a processor, updating the sparse coded dictionary with regularization using at least one auxiliary variable to provide a learned dictionary, determining whether the learned dictionary converges to an input data set, and outputting the learned dictionary regularized by the at least one a priori over-complete dictionary when the learned dictionary converges to the input data set. The system and method includes, when the learned dictionary lacks convergence, repeating the steps of performing sparse coding, updating the sparse coded dictionary, and determining whether the learned dictionary converges to the input data set.
1. A method for hierarchical sparse dictionary learning (“HiSDL”) to construct a learned dictionary regularized by an a priori over-complete dictionary, comprising:
providing at least one a priori over-complete dictionary for regularization;
performing sparse coding of the at least one a priori over-complete dictionary to provide a sparse coded dictionary;
using a processor, updating the sparse coded dictionary with regularization using auxiliary variables to provide a learned dictionary;
determining whether the learned dictionary converges to an input data set; and
outputting the learned dictionary regularized by the at least one a priori over-complete dictionary when the learned dictionary converges to the input data set.
2. The method of claim 1 , further comprising:
when the learned dictionary lacks convergence, repeating the steps of performing sparse coding, updating the sparse coded dictionary, and determining whether the learned dictionary converges to the input data set.
3. The method of claim 1 , wherein the input data set includes at least one of high-dimensional heterogeneous time series data and a data matrix.
4. The method of claim 3 , wherein the learned dictionary includes atoms from the at least one a priori over-complete dictionary which are adapted to the high-dimensional heterogeneous time series data.
5. The method of claim 1 , wherein providing the at least one a priori over-complete dictionary includes using at least one of a Discrete Cosine Transform (“DCT”) and a Haar wavelet basis to provide the at least one a priori over-complete dictionary.
6. The method of claim 1 , wherein performing the sparse coding of the at least one a priori over-complete dictionary includes using a Fast Iterative Shrinkage-Threshold (“FISTA”) approach for l 1 regularized least squares.
7. The method of claim 1 , wherein updating the sparse coded dictionary includes alternating between a first auxiliary variable at a first iteration and a subsequent auxiliary variable at a subsequent iteration.
8. The method of claim 1 , wherein updating the sparse coded dictionary further comprises eliminating obsolete atoms in the learned dictionary if any one atom in the input data set degenerates to zero.
9. The method of claim 8 , wherein any one atom in the input data set degenerates to zero if the atom does not reduce empirical error.
10. A system for hierarchical sparse dictionary learning (“HiSDL”) to construct a learned dictionary regularized by an a priori over-complete dictionary, the system comprising:
an a priori over-complete dictionary generator configured to provide at least one a priori over-complete dictionary for regularization;
a sparse coder configured to perform sparse coding of the at least one a priori over-complete dictionary to provide a sparse coded dictionary;
using a processor, a dictionary updater configured to update the sparse coded dictionary with regularization using at least one auxiliary variable to provide a learned dictionary;
a convergence determination device configured to determine convergence of the learned dictionary to an input data set; and
an output device configured to output the learned dictionary regularized by the at least one a priori over-complete dictionary when the learned dictionary converges to the input data set.
11. The system of claim 10 , wherein, when the learned dictionary lacks convergence:
the sparse coder is further configured to repeat the sparse coding;
the dictionary updater is further configured to repeat the update; and
the convergence determination device is further configured to repeat the convergence.
12. The system of claim 10 , wherein the input data set includes at least one of high-dimensional heterogeneous time series data and a data matrix.
13. The system of claim 12 , wherein the learned dictionary includes atoms from the at least one a priori over-complete dictionary which are adapted to the high-dimensional heterogeneous time series data.
14. The system of claim 10 , wherein the a priori over-complete dictionary generator is further configured to generate the at least one a priori over-complete dictionary using at least one of a Discrete Cosine Transform (“DCT”) or and a Haar wavelet basis.
15. The system of claim 10 , wherein the sparse coder is further configured to perform sparse coding of the at least one a priori over-complete dictionary using a Fast Iterative Shrinkage-Threshold (“FISTA”) approach for l 1 regularized least squares.
16. The system of claim 10 , wherein the dictionary updater is further configured to update the sparse coded dictionary by alternating between a first auxiliary variable at a first iteration and a subsequent auxiliary variable at a subsequent iteration.
17. The system of claim 10 , wherein the dictionary updater is further configured to eliminate obsolete atoms in the learned dictionary if any one atom in the input data set degenerates to zero.
18. The system of claim 17 , wherein any one atom in the input data set degenerates to zero if the atom does not reduce empirical error.
19. A computer program product is provided that includes a non-transitory computer readable storage medium having computer readable program code embodied therein for a method for hierarchical sparse dictionary learning (“HiSDL”) to construct a learned dictionary regularized by an a priori over-complete dictionary, the method comprising:
providing at least one a priori over-complete dictionary for regularization;
performing sparse coding of the at least one a priori over-complete dictionary to provide a sparse coded dictionary;
using a processor, updating the sparse coded dictionary with regularization using at least one auxiliary variables to provide a learned dictionary;
determining whether the learned dictionary converges to an input data set; and
outputting the learned dictionary regularized by the at least one a priori over-complete dictionary when the learned dictionary converges to the input data set.
20. The computer program product of claim 19 , further comprising:
when the learned dictionary lacks convergence, repeating the steps of performing sparse coding, updating the sparse coded dictionary, and determining whether the learned dictionary converges to the input data set.