Abstract
The principal components of a data matrix based on the L 2 norm can be computed with polynomial complexity via the singular-value decomposition (SVD). If, however, the principal components are constrained to be finite-alphabet or sparse or the L 1 norm is used as an alternative of the L 2 norm, then the computation of them is NP-hard. In this work, we show that in all these problems, the optimal solution can be obtained in polynomial time if the rank of the data matrix is constant. Based on the auxiliary-unit-vector technique that we have developed over the past years, we present optimal algorithms and show that they are fully parallelizable and memory efficient, hence readily implementable. We analyze the properties of our algorithms, compare against the state of the art, and comment on communications and signal processing problems where they are directly applicable to. The efficiency of our auxiliary-unit-vector technique allows the development of a binary, sparse, or L 1 principal component analysis (PCA) line of research in parallel to the conventional L 2 PCA theory.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Notes
- 1.
We ignore the other semicircle because any pair of angles ϕ 1 and ϕ 2 with difference π results in opposite vectors \(\mathbf{c}(\phi _{1}) = -\mathbf{c}(\phi _{2})\) and, hence, opposite binary vectors \(\mathbf{x}(\mathbf{c}(\phi _{1})) = -\mathbf{x}(\mathbf{c}(\phi _{2}))\) in (16) which, however, are equivalent with respect to the optimization metric in (9).
- 2.
As in the rank-2 case, we ignore the other semihypersphere because any pair of vectors \(\boldsymbol{\phi }\) and \(\tilde{\boldsymbol{\phi }}\) whose first elements ϕ 1 and \(\tilde{\phi }_{1}\), respectively, have difference π results in opposite vectors \(\mathbf{c}(\boldsymbol{\phi }) = -\mathbf{c}(\tilde{\boldsymbol{\phi }})\) and, hence, opposite binary vectors \(\mathbf{x}(\mathbf{c}(\boldsymbol{\phi })) = -\mathbf{x}(\mathbf{c}(\tilde{\boldsymbol{\phi }}))\) in (16) which, however, are equivalent with respect to the optimization metric in (9).
- 3.
If V 1: D−1, :  is full-rank, then its null space has rank 1 and \(\mathbf{c}(\boldsymbol{\phi })\) is uniquely determined (within a sign ambiguity which is resolved by c D  ≥ 0). If, instead, V 1: D−1, :  is rank-deficient, then the intersection of the D − 1 hypersurfaces (i.e., the solution of (28)) is a p-manifold (with p ≥ 1) in the (D − 1)-dimensional space and does not generate a new cell. Hence, linearly dependent combinations of D − 1 rows of V are ignored.
- 4.
An alternative way of resolving the sign ambiguities at the intersections of hypersurfaces was developed in [34] and led to the direct construction of a set S of size \(\sum _{i=0}^{D-1}\binom{N - 1}{i} = O(N^{D-1})\) with complexity O(N D).
- 5.
We again allow c(ϕ) to lie on the unit-radius semicircle and ignore the other semicircle because any pair of angles ϕ 1 and ϕ 2 with difference π results in opposite vectors \(\mathbf{c}(\phi _{1}) = -\mathbf{c}(\phi _{2})\) which, however, are equivalent with respect to the optimization metric in (40) and produce the same support \(I(\mathbf{c}(\phi _{1})) = I(\mathbf{c}(\phi _{2}))\) in (42).
- 6.
The exact detailed steps that are taken to define, with complexity O(N), the support I for each interval that is adjacent to an intersection point are described in [4].
- 7.
As in the rank-2 case, we allow \(\mathbf{c}(\boldsymbol{\phi })\) to lie on the unit-radius semihypersphere, since we can ignore the other semihypersphere because any pair of vectors \(\boldsymbol{\phi }\) and \(\tilde{\boldsymbol{\phi }}\) whose first elements ϕ 1 and \(\tilde{\phi }_{1}\), respectively, have difference π results in opposite vectors \(\mathbf{c}(\boldsymbol{\phi }) = -\mathbf{c}(\tilde{\boldsymbol{\phi }})\) which, however, are equivalent with respect to the optimization metric in (40) and produce the same support \(I(\mathbf{c}(\boldsymbol{\phi })) = I(\mathbf{c}(\tilde{\boldsymbol{\phi }}))\) in (42).
- 8.
If the (D − 1) × D matrix is full-rank, then its null space has rank 1 and \(\mathbf{c}(\boldsymbol{\phi })\) is uniquely determined (within a sign ambiguity which, however, does not affect the final decision on the index-set). If, instead, the (D − 1) × D matrix is rank-deficient, then the intersection of the D hypersurfaces (i.e., the solution of (51)) is a p-manifold (with p ≥ 1) on the D-dimensional space and does not generate a new cell. Hence, combinations of D rows of V that result in linearly dependent rows of the (D − 1) × D matrix in (51) can be simply ignored.
- 9.
Absolute-value errors put significantly less emphasis on extreme errors than squared-error expressions.
- 10.
- 11.
We note that without the presented algorithm, computation of the L 1 principal component of A 2×50 would have required complexity proportional to 250, which is of course infeasible.
References
Allemand, K., Fukuda, K., Liebling, T.M., Steiner, E.: A polynomial case of unconstrained zero-one quadratic optimization. Math. Program. A-91, 49–52 (2001)
Amini, A.A., Wainwright, M.J.: High-dimensional analysis of semidefinite relaxations for sparse principal components. In: Proceedings of IEEE ISIT 2008, pp. 2454–2458. Toronto, July 2008
d’Aspremont, A., El Ghaoui, L., Jordan, M.I., Lanckriet, G.R.G.: A direct formulation for sparse PCA using semidefinite programming. SIAM Rev. 49, 434–448 (2007)
Asteris, M., Papailiopoulos, D.S., Karystinos, G.N.: The sparse principal component of a constant-rank matrix. IEEE Trans. Inf. Theory 60, 2281–2290 (2014)
Avis, D., Fukuda, K.: Reverse search for enumeration. Discrete Appl. Math. 65, 21–46 (1996)
Barahona, F., Grötschel, M., Jünger, M., Reinelt, G.: An application of combinatorial optimization to statistical physics and circuit layout design. Oper. Res. 36, 493–513 (1988)
Baraniuk, R.G.: Compressive sensing. IEEE Signal Process. Mag. 24, 118–124 (2007)
Ben-Ameur, W., Neto, J.: A polynomial-time recursive algorithm for some unconstrained quadratic optimization problems. Discrete Appl. Math. 159, 1689–1698 (2011)
Boutsidis, C., Drineas, P., Magdon-Ismail, M.: Sparse features for PCA-like linear regression. Adv. Neural Inf. Process. Syst. 24, 2285–2293 (2011)
Brooks, J.P., Dulá, J.H.: The L1-norm best-fit hyperplane problem. Appl. Math. Lett. 26, 51–55 (2013)
Brooks, J.P., Dulá, J.H., Boone, E.L.: A pure L 1-norm principal component analysis. J. Comput. Stat. Data Anal. 61, 83–98 (2013)
Candès, E.J.: Compressive sampling. In: Proceedings of International Congress Mathematicians (ICM), pp. 1433–1452. Madrid, August 2006
Çela, E., Klinz, B., Meyer, C.: Polynomially solvable cases of the constant rank unconstrained quadratic 0–1 programming problem. J. Comb. Optim. 12, 187–215 (2006)
Cormen, T.H., Leiserson, C.E., Rivest, R.L., Stein, C.: Introduction to Algorithms, 2nd edn. MIT, Cambridge (2001)
Diederichs, E., Juditsky, A., Spokoiny, V., Schütte, C.: Sparse non-Gaussian component analysis. IEEE Trans. Inf. Theory 56, 3033–3047 (2010)
Ding, C., Zhou, D., He, X., Zha, H.: R 1-PCA: rotational invariant L 1-norm principal component analysis for robust subspace factorization. In: Proceedings of International Conference on Machine Learning Society, pp. 281–288. Pittsburgh (2006)
Dong, K., Prasad, N., Wang, X., Zhu, S.: Adaptive antenna selection and Tx/Rx beamforming for large-scale MIMO systems in 60 GHz channels. EURASIP J. Wireless Commun. Netw. (2011)
Donoho, D.L., Elad, M., Temlyakov, V.N.: Stable recovery of sparse overcomplete representations in the presence of noise. IEEE Trans. Inf. Theory 52, 6–18 (2006)
Eckart, C., Young, G.: The approximation of one matrix by another of lower rank. Psychometrika 1, 211–218 (1936)
Edelsbrunner, H.: Algorithms in Combinatorial Geometry. Springer, New York (1987)
Edelsbrunner, H., O’Rourke, J., Seidel, R.: Constructing arrangements of lines and hyperplanes with applications. SIAM J. Comput. 15, 341–363 (1986)
Eriksson, A., van den Hengel, A.: Efficient computation of robust low-rank matrix approximations in the presence of missing data using the L 1 norm. In: Proceedings of IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 771–778. San Francisco, June 2010
Ferrez, J.-A., Fukuda, K., Liebling, T.M.: Solving the fixed rank convex quadratic maximization in binary variables by a parallel zonotope construction algorithm. Eur. J. Oper. Res. 166, 35–50 (2005)
Funatsu, N., Kuroki, Y.: Fast parallel processing using GPU in computing L1-PCA bases. In: Proceedings of IEEE TENCON 2010, pp. 2087–2090. Fukuoka, November 2010
Galpin, J.S., Hawkins, D.M.: Methods of L 1 estimation of a covariance matrix. J. Comput. Stat. Data Anal. 5, 305–319 (1987)
Gieseke, F., Pahikkala, T., Igel, C.: Polynomial runtime bounds for fixed-rank unsupervised least-squares classification. In: JMLR: Workshop Conf. Proc. 29, 62–71 (2013)
Gkizeli, M., Karystinos, G.N.: Maximum-SNR antenna selection among a large number of transmit antennas. IEEE J. Select. Topics Signal Process. doi: 10.1109/JSTSP.2014.2328329 (to be published)
Golub, G.H., Van Loan, C.F.: Matrix Computations, 3rd edn. Johns Hopkins University Press, Baltimore (1996)
Grötschel, M., Jünger, M., Reinelt, G.: Via minimization with pin preassignments and layer preference. J. Appl. Math. Mech. / Zeitschrift für Angewandte Mathematik und Mechanik 69, 393–399 (1989)
Gu, Z., Lin, W., Lee, B.-S., Lau, C.T.: Rotated orthogonal transform (ROT) for motion-compensation residual coding. IEEE Trans. Image Process. 21, 4770–4781 (2012)
He, R., Hu, B.-G., Zheng, W.-S., Kong, X.-W.: Robust principal component analysis based on maximum correntropy criterion. IEEE Trans. Image Process. 20, 1485–1494 (2011)
Heath, R.W.,Jr., Paulraj, A.: A simple scheme for transmit diversity using partial channel feedback. In: Proceedings of IEEE Asilomar Conference on Signals, Systems, and Computers, vol. 2, pp. 1073–1078. Pacific Grove, November 1998
Karp, R.M.: Reducibility among combinatorial problems. In: Miller, R.E., Thatcher, J.W. (eds.) Complexity of Computer Computations, pp. 85–103. Plenum, New York (1972)
Karystinos, G.N., Liavas, A.P.: Efficient computation of the binary vector that maximizes a rank-deficient quadratic form. IEEE Trans. Inf. Theory 56, 3581–3593 (2010)
Karystinos, G.N., Pados, D.A.: Rank-2-optimal adaptive design of binary spreading codes. IEEE Trans. Inf. Theory 53, 3075–3080 (2007)
Ke, Q., Kanade, T.: Robust subspace computation using L1 norm. International Technical Report, Computer Science Department, Carnegie Mellon University, CMU-CS-03-172, August 2003
Ke, Q., Kanade, T.: Robust L 1 norm factorization in the presence of outliers and missing data by alternative convex programming. In: Proceedings of IEEE Computer Vision and Pattern Recognition (CVPR), pp. 739–746. San Diego, June 2005
Kim, Y.G., Beaulieu, N.C.: On MIMO beamforming systems using quantized feedback. IEEE Trans. Commun. 58, 820–827 (2010)
Kwak, N.: Principal component analysis based on L1-norm maximization. IEEE Trans. Pattern Anal. Mach. Intell. 30, 1672–1680 (2008)
Kwak, N., Oh, J.: Feature extraction for one-class classification problems: enhancements to biased discriminant analysis. Pattern Recog. 42, 17–26 (2009)
Kyrillidis, A., Karystinos, G.N.: Fixed-rank Rayleigh quotient maximization by an MPSK sequence. IEEE Trans. Commun. 62, 961–975 (2014)
Leung, K.-K., Sung, C.W., Khabbazian, M., Safari, M.A.: Optimal phase control for equal-gain transmission in MIMO systems with scalar quantization: complexity and algorithms. IEEE Trans. Inf. Theory 56, 3343–3355 (2010)
Li, D., Sun, X., Gu, S., Gao, J., Liu, C.: Polynomially solvable cases of binary quadratic programs. In: Chinchuluun, A., Pardalos, P.M., Enkhbat, R., Tseveendorj, I. (eds.) Optimization and Optimal Control, pp. 199–225. Springer, New York (2010)
Li, J., Petropulu, A.P.: A low complexity algorithm for collaborative-relay beamforming. In: Proceedings of IEEE ICASSP 2013, pp. 5002–5005. Vancouver, May 2013
Li, X., Pang, Y., Yuan, Y.: L1-norm-based 2DPCA. IEEE Trans. Syst. Man. Cybern. B Cybern. 40, 1170–1175 (2009)
Li, X., Hu, W., Wang, H., Zhang, Z.: Linear discriminant analysis using rotational invariant L 1 norm. Neurocomputing 73, 2571–2579 (2010)
Love, D.J., Heath, R.W., Jr., Strohmer, T.: Grassmannian beamforming for multiple-input multiple-output wireless systems. IEEE Trans. Inf. Theory 49, 2735–2747 (2003)
Luong, H.Q., Goossens, B., Aelterman, J., Pižurica, A., Philips, W.: A primal-dual algorithm for joint demosaicking and deconvolution. In: Proceedings of IEEE ICIP 2012, pp. 2801–2804, October 2012
Mackenthun, K.M., Jr.: A fast algorithm for multiple-symbol differential detection of MPSK. IEEE Trans. Commun. 42, 1471–1474 (1994)
Magnus, J.R., Neudecker, H.: Matrix Differential Calculus with Applications in Statistics and Econometrics, 2nd edn. Wiley, Chichester (1999)
Malik, U., Jaimoukha, I.M., Halikias, G.D., Gungah, S.K.: On the gap between the quadratic integer programming problem and its semidefinite relaxation. Math. Program. A-107, 505–515 (2006)
Markopoulos, P.P., Karystinos, G.N.: Novel full-rate noncoherent Alamouti encoding that allows polynomial-complexity optimal decoding. In: Proceedings of IEEE ICASSP 2013, pp. 5075–5079. Vancouver, May 2013
Markopoulos, P.P., Karystinos, G.N., Pados, D.A.: Some options for L 1-subspace signal processing. In: Proceedings of IEEE ISWCS 2013, pp. 622–626. Ilmenau, August 2013
Markopoulos, P.P., Karystinos, G.N., Pados, D.A.: Optimal algorithms for L 1-subspace signal processing. IEEE Trans. Signal Process (to be published)
McCoy, M., Tropp, J.A.: Two proposals for robust PCA using semidefinite programming. Electron. J. Stat. 5, 1123–1160 (2011)
McKilliam, R.G., Ryan, D.J., Clarkson, I.V.L., Collings, I.B.: An improved algorithm for optimal noncoherent QAM detection. In: Proceedings of 2008 Australian Communications Theory Workshop, pp. 64–68. Christchurch, Feberary 2008
Megiddo, N.: Linear programming in linear time when the dimension is fixed. J. ACM 31, 114–127 (1984)
Meng, D., Zhao, Q., Xu, Z.: Improve robustness of sparse PCA by L 1-norm maximization. Pattern Recog. 45, 487–497 (2012)
Moghaddam, B., Weiss, Y., Avidan, S.: Spectral bounds for sparse PCA: Exact and greedy algorithms. Adv. Neural Inf. Process. Syst. 18, 915–922 (2006)
Molisch, A.F., Win, M.Z.: MIMO systems with antenna selection. IEEE Microw. Mag. 5, 46–56 (2004)
Molisch, A.F., Win, M.Z., Winters, J.H.: Reduced-complexity transmit/receive-diversity systems. IEEE Trans. Signal. Process. 51, 2729–2738 (2003)
Motedayen-Aval, I., Anastasopoulos, A.: Polynomial-complexity noncoherent symbol-by-symbol detection with application to adaptive iterative decoding of turbo-like codes. IEEE Trans. Commun. 51, 197–207 (2003)
Motedayen, I., Krishnamoorthy, A., Anastasopoulos, A.: Optimal joint detection/estimation in fading channels with polynomial complexity. IEEE Trans. Inf. Theory 53, 209–223 (2007)
Nguyen, T.-D.: A fast approximation algorithm for solving the complete set packing problem. Eur. J. Oper. Res. 237, 62–70 (2014)
Nie, F., Huang, H., Ding, C., Luo, D., Wang, H.: Robust principal component analysis with non-greedy l 1-norm maximization. In: Proceedings of International Joint Conference on Artificial Intelligence (IJCAI), pp. 1433–1438. Barcelona, July 2011
Onn, S., Rothblum, U.G.: Convex combinatorial optimization. Discrete Comput. Geom. 32, 549–566 (2004)
Pang, Y., Li, X., Yuan, Y.: Robust tensor analysis with L1-norm. IEEE Trans. Circuits Syst. Video Technol. 20, 172–178 (2010)
Papailiopoulos, D.S., Karystinos, G.N.: Maximum-likelihood noncoherent OSTBC detection with polynomial complexity. IEEE Trans. Wireless Commun. 9, 1935–1945 (2010)
Papailiopoulos, D.S., Abou Elkheir, G., Karystinos, G.N.: Maximum-likelihood noncoherent PAM detection. IEEE Trans. Commun. 61, 1152–1159 (2013)
Pardalos, P.M., Rendl, F., Wolkowicz, H.: The quadratic assignment problem: a survey and recent developments. In: Proceedings of DIMACS Workshop on Quadratic Assignment Problems, vol. 16, pp. 1–42. DIMACS Series on Discrete Mathematics and Theoretical Computer Science. American Mathematical Society, Providence (1994)
Pauli, V., Lampe, L., Schober, R., Fukuda, K.: Multiple-symbol differential detection based on combinatorial geometry. IEEE Trans. Commun. 56, 1596–1600 (2008)
Pinter, R.Y.: Optimal layer assignment for interconnect. Adv. VLSI Comput. Syst. 1, 123–137 (1984)
Punnen, A., Sripratak, P., Karapetyan, D.: The bipartite unconstrained 0-1 quadratic programming problem: Polynomially solvable cases (2012). arXiv:1212.3736v3 [math.OC]. (ArXiv preprint)
Ryan, D.J., Collings, I.B., Clarkson, I.V.L.: GLRT-optimal noncoherent lattice decoding. IEEE Trans. Signal. Process. 55, 3773–3786 (2007)
Ryan, D.J., Clarkson, I.V.L., Collings, I.B., Guo, D., Honig, M.L.: QAM and PSK codebooks for limited feedback MIMO beamforming. IEEE Trans. Commun. 57, 1184–1196 (2009)
Sanayei, S., Nosratinia, A.: Antenna selection in MIMO systems. IEEE Commun. Mag. 42, 68–73 (2004)
Santipach, W., Mamat, K.: Tree-structured random vector quantization for limited-feedback wireless channels. IEEE Trans. Wireless Commun. 10, 3012–3019 (2011)
Schizas, I.D., Giannakis, G.B.: Covariance eigenvector sparsity for compression and denoising. IEEE Trans. Signal Process. 60, 2408–2421 (2012)
Shen, C., Paisitkriangkrai, S., Zhang, J.: Efficiently learning a detection cascade with sparse eigenvectors. IEEE Trans. Image Process. 20, 22–35 (2011)
Singh, N., Miller, B.A., Bliss, N.T., Wolfe, P.J.: Anomalous subgraph detection via sparse principal component analysis. In: Proceedings of IEEE SSP 2011, pp. 485–488. Nice, June 2011
Soltanalian, M., Stoica, P.: Designing unimodular codes via quadratic optimization. IEEE Trans. Signal Process. 62, 1221–1234 (2014)
Sung, C.W., Kwan, H.Y.: Heuristic algorithms for binary sequence assignment in DS-CDMA systems. In: Proceedings of IEEE PIMRC 2002, vol. 5, pp. 2327–2331. Lisbon, September 2002
Sweldens, W.: Fast block noncoherent decoding. IEEE Commun. Lett. 5, 132–134 (2001)
Tropp, J.A.: Just relax: convex programming methods for identifying sparse signals in noise. IEEE Trans. Inf. Theory 52, 1030–1051 (2006)
Ulfarsson, M.O., Solo, V.: Vector l 0 sparse variable PCA. IEEE Trans. Signal Process. 59, 1949–1958 (2011)
Van Trees, H.L.: Detection, Estimation, and Modulation Theory (Part I). Wiley, Hoboken (2001)
Wang, H.: Block principal component analysis with L1-norm for image analysis. Pattern Recog. Lett. 33, 537–542 (2012)
Wang, H., Tang, Q., Zheng, W.: L1-norm-based common spatial patterns. IEEE Trans. Biomed. Eng. 59, 653–662 (2012)
Wei, D., Sestok, C.K., Oppenheim, A.V.: Sparse filter design under a quadratic constraint: low-complexity algorithms. IEEE Trans. Signal Process. 61, 857–870 (2013)
Wong, T.F., Lok, T.M.: Transmitter adaptation in multicode DS-CDMA systems. IEEE J. Select. Areas Commun. 19, 69–82 (2001)
Wu, M., Kam, P.Y.: Sequence detection on fading channels without explicit channel estimation. In: Proceedings of IEEE Wireless VITAE 2009, pp. 370–374. Aalborg, May 2009
Wu, M., Kam, P.Y.: Performance analysis and computational complexity comparison of sequence detection receivers with no explicit channel estimation. IEEE Trans. Vehic. Tech. 59, 2625–2631 (2010)
Xu, W., Stojnic, M., Hassibi, B.: Low-complexity blind maximum-likelihood detection for SIMO systems with general constellations. In: Proceedings of IEEE ICASSP 2008, pp. 2817–2820. Las Vegas, April 2008
Xu, M., Guo, D., Honig, M.L.: MIMO precoding with limited rate feedback: simple quantizers work well. In: Proceedings of IEEE GLOBECOM 2009. Honolulu, December 2009
Yu, L., Zhang, M., Ding, C.: An efficient algorithm for L1-norm principal component analysis. In: Proceedings of IEEE ICASSP 2012, pp. 1377–1380. Kyoto, March 2012
Zheng, X., Xie, Y., Li, J., Stoica, P.: MIMO transmit beamforming under uniform elemental power constraint. IEEE Trans. Signal Process. 55, 5395–5406 (2007)
Zou, H., Hastie, T., Tibshirani, R.: Sparse principal component analysis. J. Comput. Graph. Stat. 15, 265–286 (2006)
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2014 Springer Science+Business Media New York
About this chapter
Cite this chapter
Karystinos, G.N. (2014). Optimal Algorithms for Binary, Sparse, and L 1-Norm Principal Component Analysis. In: Pardalos, P., Rassias, T. (eds) Mathematics Without Boundaries. Springer, New York, NY. https://doi.org/10.1007/978-1-4939-1124-0_11
Download citation
DOI: https://doi.org/10.1007/978-1-4939-1124-0_11
Published:
Publisher Name: Springer, New York, NY
Print ISBN: 978-1-4939-1123-3
Online ISBN: 978-1-4939-1124-0
eBook Packages: Mathematics and StatisticsMathematics and Statistics (R0)