Acoustics, Speech, and Signal Processing, IEEE International Conference on (2000)
June 5, 2000 to June 9, 2000
P.A. Regalia , Inst. Nat. des Telecommun., Evry, France
We revisit the higher-order power method of De Lathauwer et al. (1995) for rank-one tensor approximation, and its relation to contrast maximization as used in blind deconvolution. We establish a simple convergence proof for the general nonsymmetric tensor case. We show also that a symmetric version of the algorithm, offering an order of magnitude reduction in computational complexity but discarded by De Lathauwer et al. as unpredictable, is likewise provably convergent. A new initialization scheme is also developed which, unlike the TSVD-based initialization, leads to a quantifiable proximity to the globally optimal solution.
E. Kofidis and P. Regalia, "The higher-order power method revisited: convergence proofs and effective initialization," Acoustics, Speech, and Signal Processing, IEEE International Conference on(ICASSP), Istanbul, Turkey, 2000, pp. 2709-2712.