Title :
Natural gradient for minor component extraction
Author :
Hasan, Mohammed A.
Author_Institution :
Dept. of Electr. & Comput. Eng., Minnesota Univ., Duluth, MN, USA
Abstract :
The paper proposes constrained optimization criteria for extracting in parallel multiple minor components using a weighted inverse Rayleigh quotient (WIRQ). This WIRQ is formulated into many constrained optimization problems which results in deriving many variations of MCA flows using the natural gradient concept. The derived minor component flows are analyzed and some of their convergence properties are presented. Procedures for converting MCA flows into a principal component flow are also discussed.
Keywords :
data analysis; eigenvalues and eigenfunctions; gradient methods; optimisation; principal component analysis; signal processing; constrained optimization problems; data analysis; eigenvalues; minor component analysis; minor component extraction; natural gradient; principal component analysis; principal component flow; signal processing; weighted inverse Rayleigh quotient; Concurrent computing; Constraint optimization; Convergence; Cost function; Data analysis; Eigenvalues and eigenfunctions; Principal component analysis; Signal analysis; Signal processing algorithms; Symmetric matrices;
Conference_Titel :
Circuits and Systems, 2005. ISCAS 2005. IEEE International Symposium on
Print_ISBN :
0-7803-8834-8
DOI :
10.1109/ISCAS.2005.1465791
