a new method for searching an l1 solution of an...
TRANSCRIPT
LAD ProblemBasic properties
New methodAplication
A new method for searching an L1 solution of anoverdetermined system of linear equations and
applications
Goran Kusec1 Ivana Kuzmanovic2 Kristian Sabo2
Rudolf Scitovski2
1Faculty of Agriculture, University of Osijek2Department of Mathematics, University of Osijek
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
LAD Problem
A ∈ Rm×n, (m� n), b ∈ Rm
A =
? ? . . . ?? ? . . . ?? ? . . . ?? ? . . . ?...
......
......
......
...? ? . . . ?? ? . . . ?? ? . . . ?? ? . . . ?
︸ ︷︷ ︸
, b =
????......????
m
n
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
LAD Problem
A ∈ Rm×n, (m� n), b ∈ Rm
A =
? ? . . . ?? ? . . . ?? ? . . . ?? ? . . . ?...
......
......
......
...? ? . . . ?? ? . . . ?? ? . . . ?? ? . . . ?
︸ ︷︷ ︸
, b =
????......????
m
n
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
system of linear equations Ax = b
b ∈ R(A)
b R(A)
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
overdetermined system of linear equations Ax ≈ b
b /∈ R(A)
b
R(A)
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
‖b− Ax‖ → minx∈Rn
Euclidean l2-norm (Least Square (LS) solution)l1-norm (Least absolute deviation (LAD) solution)l∞-norm (Tschebishev solution)
applications in various fields of applied researchJ. A. Cadzow, Minimum l1, l2 and l∞ Norm Approximate Solutions to an Overdetermined System
of Linear Equations, Digital Signal Processing 12(2002) 524–560
S. Dasgupta, S. K. Mishra, Least absolute deviation estimation of linear econometric models: A
literature review, Munich Personal RePEc Archive, 1-26, 2004
C. Gurwitz, Weighted median algorithms for L1 approximation, BIT 30(1990) 301–310
Dodge Y. (Ed.), Statistical Data Analysis Based on the L1-norm and Related Methods, Proceedingsof The Thrid International Conference on Statistical Data Analysis Based on the L1-norm andRelated Methods, Elsevier, Neuchatel, 1997
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
‖b− Ax‖ → minx∈Rn
Euclidean l2-norm (Least Square (LS) solution)l1-norm (Least absolute deviation (LAD) solution)l∞-norm (Tschebishev solution)
applications in various fields of applied researchJ. A. Cadzow, Minimum l1, l2 and l∞ Norm Approximate Solutions to an Overdetermined System
of Linear Equations, Digital Signal Processing 12(2002) 524–560
S. Dasgupta, S. K. Mishra, Least absolute deviation estimation of linear econometric models: A
literature review, Munich Personal RePEc Archive, 1-26, 2004
C. Gurwitz, Weighted median algorithms for L1 approximation, BIT 30(1990) 301–310
Dodge Y. (Ed.), Statistical Data Analysis Based on the L1-norm and Related Methods, Proceedingsof The Thrid International Conference on Statistical Data Analysis Based on the L1-norm andRelated Methods, Elsevier, Neuchatel, 1997
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
‖b− Ax‖ → minx∈Rn
Euclidean l2-norm (Least Square (LS) solution)l1-norm (Least absolute deviation (LAD) solution)l∞-norm (Tschebishev solution)
applications in various fields of applied researchJ. A. Cadzow, Minimum l1, l2 and l∞ Norm Approximate Solutions to an Overdetermined System
of Linear Equations, Digital Signal Processing 12(2002) 524–560
S. Dasgupta, S. K. Mishra, Least absolute deviation estimation of linear econometric models: A
literature review, Munich Personal RePEc Archive, 1-26, 2004
C. Gurwitz, Weighted median algorithms for L1 approximation, BIT 30(1990) 301–310
Dodge Y. (Ed.), Statistical Data Analysis Based on the L1-norm and Related Methods, Proceedingsof The Thrid International Conference on Statistical Data Analysis Based on the L1-norm andRelated Methods, Elsevier, Neuchatel, 1997
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
Vector x∗ ∈ Rn such that
minx∈Rn‖b− Ax‖1 = ‖b− Ax∗‖1
is called the best LAD-solution of overdetermined system Ax ≈ b.
operational research literature - a parameter estimationproblem of a hyperplane on the basis of a given set of points
N. M. Korneenko, H. Martini, Hyperplane approximation and related topics, in: New Trends inDiscrete and Computational Geometry, (J. Pach, Ed.), Springer-Verlag, Berlin, 1993.
A. Schobel, Locating Lines and Hyperplanes: Theory and Algorithms, Springer Verlag, Berlin, 1999.
statistics literature - a parameter estimation problem for linearregression
S. Dasgupta, S. K. Mishra, Least absolute deviation estimation of linear econometric models: A
literature review, Munich Personal RePEc Archive, 1-26, 2004
E. Z. Demidenko, Optimization and Regression, Nauka, Moscow, 1989, in Russian
Dodge Y. (Ed.) (1997): Statistical Data Analysis Based on the L1-norm and Related Methods,Proceedings of The Thrid International Conference on Statistical Data Analysis Based on theL1-norm and Related Methods, Elsevier, Neuchatel
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
Vector x∗ ∈ Rn such that
minx∈Rn‖b− Ax‖1 = ‖b− Ax∗‖1
is called the best LAD-solution of overdetermined system Ax ≈ b.
operational research literature - a parameter estimationproblem of a hyperplane on the basis of a given set of points
N. M. Korneenko, H. Martini, Hyperplane approximation and related topics, in: New Trends inDiscrete and Computational Geometry, (J. Pach, Ed.), Springer-Verlag, Berlin, 1993.
A. Schobel, Locating Lines and Hyperplanes: Theory and Algorithms, Springer Verlag, Berlin, 1999.
statistics literature - a parameter estimation problem for linearregression
S. Dasgupta, S. K. Mishra, Least absolute deviation estimation of linear econometric models: A
literature review, Munich Personal RePEc Archive, 1-26, 2004
E. Z. Demidenko, Optimization and Regression, Nauka, Moscow, 1989, in Russian
Dodge Y. (Ed.) (1997): Statistical Data Analysis Based on the L1-norm and Related Methods,Proceedings of The Thrid International Conference on Statistical Data Analysis Based on theL1-norm and Related Methods, Elsevier, Neuchatel
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
principle is considered to have been proposed by the Croatianmathematician J. R. Boskovic in the mid-eighteenth century
D. Birkes, Y. Dodge, Alternative Methods of Regression, Wiley, New York, 1993
P. Bloomfield, W. Steiger, Least Absolute Deviations: Theory, Applications, and Algorithms,Birkhauser, Boston, 1983
Dodge Y. (Ed.) (1997): Statistical Data Analysis Based on the L1-norm and Related Methods,Proceedings of The Thrid International Conference on Statistical Data Analysis Based on theL1-norm and Related Methods, Elsevier, Neuchatel
Numerical methodsclassical nondifferentiable minimization methods cannot beapplied directly-unreasonably long computing timespecialized algorithms
J. A. Cadzow, Minimum l1, l2 and l∞ Norm Approximate Solutions to an Overdetermined System of Linear
Equations, Digital Signal Processing 12(2002) 524–560
E. Castillo, R. Mınguez, C. Castillo, A. S. Cofinno, Dealing with the multiplicity of solutions of the l1 and l∞regression models, European J. Oper. Res. 188(2008), 460–484.
Y. Li, A maximum likelihood approach to least absolute deviation regression, EURASIP Journal on Applied
Signal Processing 12(2004), 1762–1769
N. T. Trendafilov, G. A. Watson, The l1 oblique procrustes problem, Statistics and Computing 14(2004)
39–51
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
principle is considered to have been proposed by the Croatianmathematician J. R. Boskovic in the mid-eighteenth century
D. Birkes, Y. Dodge, Alternative Methods of Regression, Wiley, New York, 1993
P. Bloomfield, W. Steiger, Least Absolute Deviations: Theory, Applications, and Algorithms,Birkhauser, Boston, 1983
Dodge Y. (Ed.) (1997): Statistical Data Analysis Based on the L1-norm and Related Methods,Proceedings of The Thrid International Conference on Statistical Data Analysis Based on theL1-norm and Related Methods, Elsevier, Neuchatel
Numerical methodsclassical nondifferentiable minimization methods cannot beapplied directly-unreasonably long computing timespecialized algorithms
J. A. Cadzow, Minimum l1, l2 and l∞ Norm Approximate Solutions to an Overdetermined System of Linear
Equations, Digital Signal Processing 12(2002) 524–560
E. Castillo, R. Mınguez, C. Castillo, A. S. Cofinno, Dealing with the multiplicity of solutions of the l1 and l∞regression models, European J. Oper. Res. 188(2008), 460–484.
Y. Li, A maximum likelihood approach to least absolute deviation regression, EURASIP Journal on Applied
Signal Processing 12(2004), 1762–1769
N. T. Trendafilov, G. A. Watson, The l1 oblique procrustes problem, Statistics and Computing 14(2004)
39–51
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
principle is considered to have been proposed by the Croatianmathematician J. R. Boskovic in the mid-eighteenth century
D. Birkes, Y. Dodge, Alternative Methods of Regression, Wiley, New York, 1993
P. Bloomfield, W. Steiger, Least Absolute Deviations: Theory, Applications, and Algorithms,Birkhauser, Boston, 1983
Dodge Y. (Ed.) (1997): Statistical Data Analysis Based on the L1-norm and Related Methods,Proceedings of The Thrid International Conference on Statistical Data Analysis Based on theL1-norm and Related Methods, Elsevier, Neuchatel
Numerical methodsclassical nondifferentiable minimization methods cannot beapplied directly-unreasonably long computing timespecialized algorithms
J. A. Cadzow, Minimum l1, l2 and l∞ Norm Approximate Solutions to an Overdetermined System of Linear
Equations, Digital Signal Processing 12(2002) 524–560
E. Castillo, R. Mınguez, C. Castillo, A. S. Cofinno, Dealing with the multiplicity of solutions of the l1 and l∞regression models, European J. Oper. Res. 188(2008), 460–484.
Y. Li, A maximum likelihood approach to least absolute deviation regression, EURASIP Journal on Applied
Signal Processing 12(2004), 1762–1769
N. T. Trendafilov, G. A. Watson, The l1 oblique procrustes problem, Statistics and Computing 14(2004)
39–51
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
LAD problem ⇔
On the basis of the given set of points
Λ = {Ti (x(i)1 , . . . , x
(i)n−1, z
(i)) ∈ Rn : i ∈ I}, I = {1, . . . ,m}, vectora∗ = (a∗0, a
∗1, . . . , a
∗n−1) ∈ Rn of optimal parameters of hyperplane
f (x; a) = a0 +∑n−1
j=1 ajxj should be determined such that:
G (a∗) = mina∈Rn
G (a), G (a) =m∑
i=1
∣∣∣∣∣∣z(i) − a0 −n−1∑j=1
ajx(i)j
∣∣∣∣∣∣ .
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
n = 2
Data-points Λ = {(x (i)1 , z(i)), i = 1, . . . ,m}
LAD line z = a∗0 + a∗1x1
min(a0,a1)∈R2
m∑i=1
|z(i) − a0 − a1x(i)1 | =
m∑i=1
|z(i) − a∗0 − a∗1x(i)1 |
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
n = 2
Data-points Λ = {(x (i)1 , z(i)), i = 1, . . . ,m}
LAD line z = a∗0 + a∗1x1
min(a0,a1)∈R2
m∑i=1
|z(i) − a0 − a1x(i)1 | =
m∑i=1
|z(i) − a∗0 − a∗1x(i)1 |
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
n = 2
Data-points Λ = {(x (i)1 , z(i)), i = 1, . . . ,m}
LAD line z = a∗0 + a∗1x1
min(a0,a1)∈R2
m∑i=1
|z(i) − a0 − a1x(i)1 | =
m∑i=1
|z(i) − a∗0 − a∗1x(i)1 |
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
n = 2
Data-points Λ = {(x (i)1 , z(i)), i = 1, . . . ,m}
LAD line z = a∗0 + a∗1x
min(a0,a1)∈R2
m∑i=1
|z(i) − a0 − a1x(i)1 | =
m∑i=1
|z(i) − a∗0 − a∗1x(i)1 |
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
LAD solution always exists
N. M. Korneenko, H. Martini, Hyperplane approximation and related topics, in: New Trends in Discrete andComputational Geometry, (J. Pach, Ed.), Springer-Verlag, Berlin, 1993.
A. Pinkus, On L1-approximation, Cambridge University Press, New York, 1993.
G. A. Watson, Approximation Theory and Numerical Methods, John Wiley & Sons, Chichester, 1980.
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
LAD solution is generally not unique
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
n=2, LAD solution is not unique
0.2 0.4 0.6 0.8 1
0.2
0.4
0.6
0.8
1
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
n=2, LAD solution is not unique
0.2 0.4 0.6 0.8 1
0.2
0.4
0.6
0.8
1
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
n=2, LAD solution is not unique
0.2 0.4 0.6 0.8 1
0.2
0.4
0.6
0.8
1
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
n=2, LAD solution is not unique
0.2 0.4 0.6 0.8 1
0.2
0.4
0.6
0.8
1
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
n=2, LAD solution is not unique
0.2 0.4 0.6 0.8 1
0.2
0.4
0.6
0.8
1
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
n=2, LAD solution is not unique
0.2 0.4 0.6 0.8 1
0.2
0.4
0.6
0.8
1
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
n=2, LAD solution is not unique
0.2 0.4 0.6 0.8 1
0.2
0.4
0.6
0.8
1
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
n=2, LAD solution is not unique
0.2 0.4 0.6 0.8 1
0.2
0.4
0.6
0.8
1
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
n=2, LAD solution is not unique
0.2 0.4 0.6 0.8 1
0.2
0.4
0.6
0.8
1
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
Definition
Matrix A ∈ Rm×n (m ≥ n) is said to satisfy the Haar condition ifevery n × n submatrix is nonsingular.
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
Theorem 1
Let A ∈ Rm×n, m > n, be a matrix of full column rank andb = (b1, . . . , bm)T ∈ Rm a given vector. Then there exists a permutationmatrix Π ∈ Rm×m such that
ΠA =
[A1
A2
], Πb =
[b1
b2
], A1 ∈ Rn×n, A2 ∈ R(m−n)×n, b1 ∈ Rn, b2 ∈ Rm−n,
whereby A1 is a nonsingular matrix, and there exists a LAD-solution x∗ ∈ Rn
such that A1x∗ = b1.Furthermore, if matrix [A b] satisfies the Haar condition, then the solutionx∗ ∈ Rn of the system A1x = b1 is a solution of LAD problem if and only ifvector
v = (A1T )−1A2
T s, si := sign (b2 − A2x∗)i
satisfies ‖v‖∞ ≤ 1.
Furthermore, x∗ is a unique solution of LAD problem if and only if ‖v‖∞ < 1.
A. Pinkus, On L1-approximation, Cambridge University Press, New York, 1993.
G. A. Watson, Approximation Theory and Numerical Methods, John Wiley & Sons, Chichester, 1980.
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
parameter estimation problem of a hyperplane for a set ofpoints in a plane
n = 2 for the given set of points in a plane, with condition thatnot all of them lie on some line parallel to the second axis,there exists the best LAD-line passing through at least twodifferent data pointsn = 3 for the given set of points in the space, with conditionthat not all of them lie on some plane parallel to the third axis,there exists the best LAD-plane passing through at least threedifferent data points
Y. Li, A maximum likelihood approach to least absolute deviation regression, EURASIP Journal on Applied
Signal Processing 12(2004), 1762–1769
R. Scitovski, K. Sabo, I. Kuzmanovic, I. Vazler, R. Cupec, R. Grbic, Three points method for searching the
best least absolute deviations plane, 4th Croatian Mathematical Congress, Osijek, June 17 – 20, 2008
A. Schobel, Locating Lines and Hyperplanes: Theory and Algorithms, Springer Verlag, Berlin, 1999.
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
parameter estimation problem of a hyperplane for a set ofpoints in a plane
n = 2 for the given set of points in a plane, with condition thatnot all of them lie on some line parallel to the second axis,there exists the best LAD-line passing through at least twodifferent data pointsn = 3 for the given set of points in the space, with conditionthat not all of them lie on some plane parallel to the third axis,there exists the best LAD-plane passing through at least threedifferent data points
Y. Li, A maximum likelihood approach to least absolute deviation regression, EURASIP Journal on Applied
Signal Processing 12(2004), 1762–1769
R. Scitovski, K. Sabo, I. Kuzmanovic, I. Vazler, R. Cupec, R. Grbic, Three points method for searching the
best least absolute deviations plane, 4th Croatian Mathematical Congress, Osijek, June 17 – 20, 2008
A. Schobel, Locating Lines and Hyperplanes: Theory and Algorithms, Springer Verlag, Berlin, 1999.
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
New method
I. Kuzmanovic, K. Sabo, R. Scitovski, Method for searching the best least absolute deviations solution of an
overdetermined system of linear equations, submitted
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
numerical methods for searching the best LAD-solution:Hooke and Jeeves Method, Differential Evolution, NelderMead, Random Search, Simulated Annealing
J. A. Nelder, R. Mead, A simplex method for function minimization, Comput. J. 7(1965)
308–313
C. T. Kelley, Iterative Methods for Optimization, SIAM, Philadelphia, 1999.
various specializations of the Gauss-Newton method
E. Z. Demidenko, Optimization and Regression, Nauka, Moscow, 1989, in Russia
M. R. Osborne, Finite Algorithms in Optimization and Data Analysis, Wiley, Chichester,1985.
Linear Programming methods:
A. Barrodale, F. D. K. Roberts, An improved algorithm for discrete l1 linear approximation,
SIAM J. Numer. Anal. 10(1973) 839–848
N. M. Korneenko, H. Martini, Hyperplane approximation and related topics, in: New Trendsin Discrete and Computational Geometry, (J. Pach, Ed.), Springer-Verlag, Berlin, 1993.
some of specialized methods: Barrodale-Roberts,Bartels-Conn-Sinclair, Bloomfield-Steiger
N. T. Trendafilov, G. A. Watson, The l1 oblique procrustes problem, Statistics and
Computing 14(2004) 39–51
P. Bloomfield, W. Steiger, Least Absolute Deviations: Theory, Applications, and Algorithms,Birkhauser, Boston, 1983
Y. Dodge, Statistical Data Analysis Based on the L1-norm and Related Methods,Proceedings of The Thrid International Conference on Statistical Data Analysis Based onthe L1-norm and Related Methods, Elsevier, Neuchatel, 1997
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
Lemma 1
Let y1 ≤ y2 ≤ . . . ≤ ym be real numbers with corresponding weights wi > 0,W :=
∑mi=1 wi , and I = {1, . . . ,m}, m ≥ 2 a set of indices. Denote
ν0 =
{max
{ν ∈ I :
∑νi=1 wi ≤ W
2
}, if
{ν ∈ I :
∑νi=1 wi ≤ W
2
}6= ∅
0, otherwise
Furthermore, let F : R→ R be a function defined by the formulaF (α) =
∑mi=1 wi |yi − α|. Then
(i) if∑ν0
i=1 wi <W2
, then the minimum of the function F is attained at thepoint α? = yν0+1;
(ii) if∑ν0
i=1 wi = W2
, then the minimum of the function F is attained at everypoint α? from the segment [yν0 , yν0+1].
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
Lemma 2
Let x0 be a unique solution of the system Ax = b, where A ∈ Rn×n is a squarenonsingular matrix and b ∈ Rn a given vector.If the j-th equation of this system aT
j x = bj is replaced with equation uT x = β,whereby uT dj 6= 0, where dj ∈ Rn is the j-th column of matrix A−1, then thesolution of a new system is given by
x = x0 + αdj , α =β − uT x0
uT dj.
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
Definition
Vector x∗Jk∈ Rn, Jk = {i1, . . . , ik} ⊆ I , k ≤ n is called the best
Jk -LAD-solution of problem
minx∈Rn‖b− Ax‖1 = ‖b− Ax∗‖1
if on x∗Jkthe minimum of functional x 7→ ‖r(x)‖1 is attained with
conditionri (x∗Jk
) := bi − (Ax)i = 0, i ∈ Jk .
the best Jn-LAD-solution is obtained by solving the system oflinear equations
ri (x) = 0, i = i1, . . . , in,
the best Jn−1-LAD-solution is obtained as a solution of oneweighted median problem.
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
Lemma 3
Let A ∈ Rm×n , (m > n) be a matrix of full column rank, b ∈ Rm a given vector and e = (1, . . . ,m)T a vector
of indices. Furthermore, let Π ∈ Rm×m be a permutation matrix such that
ΠA =
[A−1A+
2
]=
[A11 a1k A12A21 a2k A22
], Πb =
[b−1b+
2
],
where A11 ∈ R(n−1)×(k−1), A12 ∈ R(n−1)×(n−k), A21 ∈ R(m−n+1)×(k−1), A22 ∈ R(m−n+1)×(n−k),
a1k , b−1 ∈ Rn−1, a2k , b+2 ∈ Rm−n+1, whereby index k is such that A1k := [A11 A12] is a nonsingular matrix.
Let A2k := [A21 A22], I− = {(Πe)k : k = 1, . . . , n − 1} and I + = {(Πe)k : k = n, n + 1, . . . ,m}.
Then there exists i0 ∈ I + and the best Jn-LAD-solution x∗ ∈ Rn , such that ri (x∗) = 0 for all i ∈ I− ∪ {i0}.The k-th component x∗k of vector x∗ is thereby a solution of the weighted median problem
‖A2k η − b+2 − α(A2k ξ − a2k )‖1 → min
α, (1)
whereby ξ, i.e. η, is a unique solution of systems
A1k ξ = a1k , i.e. A1k η = b−1 , (2)
whereas other components of vector x∗ are obtained by solving the system
A1k (x1, . . . , xk−1, xk+1, . . . , xn)T = b−1 − x∗k a1k . (3)
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
matrices A and [A b] - satisfy the Haar condition.
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
Step 0 (Input) m, n, e = (1, . . . ,m)T , A ∈ Rm×n, b ∈ Rm , Π0 ∈ Rm×m .
Step 1 (Defining the first submatrix) Determine index k such that A1k := [A11 A12] ∈ R(n−1)×(n−1) is anonsingular matrix, where
H0 := Π0A =
[A−1A+
2
]=
[A11 a1k A12A21 a2k A22
], Π0b =
[b−1b+
2
].
Set I + = {(Π0e)i : i = n, n + 1, . . . ,m} and e := Π0e.
Step 2 (Searching for a new equation) According to Lemma 3, determine i0 ∈ I + and solution x(0). Calculate
G0 = G(x(0)) := ‖b− Ax(0)‖1
Let A1 =
[A−1aT
i0
]∈ Rn×n , b1 =
[b−1bi0
]∈ Rn .
Furthermore, let A2 ∈ R(m−n)×n be a matrix obtained from A+2 by dropping the i0-th row, b2 ∈ Rm−n
a vector obtained from b+2 by dropping the i0-th component
Solve system
AT1 v = AT
2 s, si := sign (b2 − A2x)i , i = 1, . . . , n,
and denote the solution by v(0)
Determine j0 such that |v (0)j0| = max
i=1,...,n|v (0)
i | =: vM
If vM ≤ 1, STOP; otherwise go to Step 3
Step 3 (Equations exchange) Let Π1 ∈ Rm×m be a permutation matrix which in matrix H0 replaces the j0-th
row with the i0-th row. Determine index k such that A1k := [A11 A12] ∈ R(n−1)×(n−1) is a nonsingularmatrix, where
H1 := Π1H0 =
[A−1A+
2
]=
[A11 a1k A12A21 a2k A22
], Π1b =
[b−1b+
2
],
Set I + = {(Π1e)i : i = n, n + 1, . . . ,m} and e := Π1e.
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
Step 4 (Searching for a new equation) According to Lemma 2, determine i0 ∈ I + and solution x(1). Calculate
G1 = G(x(1)) := ‖b− Ax(1)‖1;
Step 5 (Preparation for a new loop) Let A1 =
[A−1aT
i0
]∈ Rn×n , b1 =
[b−1bi0
]∈ Rn .
Furthermore, let A2 ∈ R(m−n)×n be a matrix obtained from A+2 by dropping the i0-th row, b2 ∈ Rm−n
a vector obtained from b+2 by dropping the i0-th component.
Solve system
AT1 v = AT
2 s, si := sign (b2 − A2x)i , i = 1, . . . , n,
and denote the solution by v(1)
Determine j0 such that |v (1)j0| = max
i=1,...,n|v (1)
i | =: vM
If vM ≤ 1, STOP; otherwise put H0 = H1 and go to Step 3
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
Initialization
The initial matrix A−1matrix A satisfies the Haar conditionthe first (n − 1) rows of matrix AΠ0 = I- identity matrix
A nonsingular submatrix A1k of matrix
H0 := Π0A =
[A−1A+
2
]=
[A11 a1k A12
A21 a2k A22
]in Step 1 - by applying QR factorization with column pivoting
A−1 Π := [A11 a1k A12] Π = Q[R ρ],
Π ∈ Rn×n - permutation matrixQ ∈ R(n−1)×(n−1) - an orthogonal matrixR ∈ R(n−1)×(n−1) - an upper triangular nonsingular matrixρ ∈ R(n−1)
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
Initialization
The initial matrix A−1matrix A satisfies the Haar conditionthe first (n − 1) rows of matrix AΠ0 = I- identity matrix
A nonsingular submatrix A1k of matrix
H0 := Π0A =
[A−1A+
2
]=
[A11 a1k A12
A21 a2k A22
]in Step 1 - by applying QR factorization with column pivoting
A−1 Π := [A11 a1k A12] Π = Q[R ρ],
Π ∈ Rn×n - permutation matrixQ ∈ R(n−1)×(n−1) - an orthogonal matrixR ∈ R(n−1)×(n−1) - an upper triangular nonsingular matrixρ ∈ R(n−1)
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
A1k := [A11 A12] = Q R ΠT1 , a1k = Qρ,
Π1 is the matrix - from Π by dropping the kth row and thenth column.
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
Theorem 2
Let x be approximation of LAD-problem obtained in Step 2, i.e. Step 4, of the
Algorithm as a solution of the system A1x = b1, where A1 =
[A−1aT
i0
]∈ Rn×n,
b1 =
[b−1bi0
]∈ Rn. Furthermore, let A2 ∈ R(m−n)×n be a matrix obtained from
A+2 by dropping the i0-th row, b2 ∈ Rm−n a vector obtained from b+
2 bydropping the i0-th component, and v ∈ Rn a solution of the system
AT1 v = AT
2 s, si := sign (b2 − A2x)i , i = 1, . . . , n.
Denote|vj0 | = max
i=1,...,n|vi |.
I. If |vj0 | ≤ 1, then x is the best LAD-solution;
II. If |vj0 | > 1, then j0 6= n, and maximum decreasing of minimizing functionvalues is attained such that in Step 3 the j0-th row of matrix A1 isreplaced.
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
Convergence Theorem 4
Let A ∈ Rm×n be a matrix and b ∈ Rm a given vector, m ≥ n,such that matrices A and [A b] satisfy the Haar condition. Thensequence (x(n)) defined by the iterative procedure in the Algorithmconverges in finitely many steps to the best LAD-solution.
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
Example 1
Table 1.
i A b1 42 7 -28 34 13 -49 62 31 46 -31 -12 17 18 -423 13 -11 -1 -35 31 -33 484 45 -35 43 -9 -36 -8 -165 25 -34 37 -30 -2 32 136 -37 -24 -9 -35 -35 21 -167 -12 26 -49 -49 -11 -22 -328 -42 31 28 -5 -45 47 -289 19 2 -29 -33 0 -46 35
10 -21 -26 -45 -44 5 26 4711 28 26 -38 -39 -8 -16 -612 32 27 -7 9 -43 -22 -38
G0 = 236.346Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
Example 1
Table 1.
i A b1 42 7 -28 34 13 -49 62 31 46 -31 -12 17 18 -423 13 -11 -1 -35 31 -33 484 45 -35 43 -9 -36 -8 -165 25 -34 37 -30 -2 32 136 -37 -24 -9 -35 -35 21 -167 -12 26 -49 -49 -11 -22 -328 -42 31 28 -5 -45 47 -289 19 2 -29 -33 0 -46 35
10 -21 -26 -45 -44 5 26 4711 28 26 -38 -39 -8 -16 -612 32 27 -7 9 -43 -22 -38
G1 = 180.613Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
Example 1
Table 1.
i A b1 42 7 -28 34 13 -49 62 31 46 -31 -12 17 18 -423 13 -11 -1 -35 31 -33 484 45 -35 43 -9 -36 -8 -165 25 -34 37 -30 -2 32 136 -37 -24 -9 -35 -35 21 -167 -12 26 -49 -49 -11 -22 -328 -42 31 28 -5 -45 47 -289 19 2 -29 -33 0 -46 35
10 -21 -26 -45 -44 5 26 4711 28 26 -38 -39 -8 -16 -612 32 27 -7 9 -43 -22 -38
G2 = 171.134Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
Example 1
Table 1.
i A b1 42 7 -28 34 13 -49 62 31 46 -31 -12 17 18 -423 13 -11 -1 -35 31 -33 484 45 -35 43 -9 -36 -8 -165 25 -34 37 -30 -2 32 136 -37 -24 -9 -35 -35 21 -167 -12 26 -49 -49 -11 -22 -328 -42 31 28 -5 -45 47 -289 19 2 -29 -33 0 -46 35
10 -21 -26 -45 -44 5 26 4711 28 26 -38 -39 -8 -16 -612 32 27 -7 9 -43 -22 -38
G3 = 148.12Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
Example 1
Table 1.
i A b1 42 7 -28 34 13 -49 62 31 46 -31 -12 17 18 -423 13 -11 -1 -35 31 -33 484 45 -35 43 -9 -36 -8 -165 25 -34 37 -30 -2 32 136 -37 -24 -9 -35 -35 21 -167 -12 26 -49 -49 -11 -22 -328 -42 31 28 -5 -45 47 -289 19 2 -29 -33 0 -46 35
10 -21 -26 -45 -44 5 26 4711 28 26 -38 -39 -8 -16 -612 32 27 -7 9 -43 -22 -38
G4 = 146.663Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
Application
Data-points Ti = (x(i)1 , x
(i)2 , z(i)), i = 1, . . . ,m
x(i)1 -fat thickness
x(i)2 -lumbar muscle thickness
z (i)- muscle percent of the i th pig
z = a0 + a1x1 + a2x2
outliers
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
Application
Data-points Ti = (x(i)1 , x
(i)2 , z(i)), i = 1, . . . ,m
x(i)1 -fat thickness
x(i)2 -lumbar muscle thickness
z (i)- muscle percent of the i th pig
z = a0 + a1x1 + a2x2
outliers
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
Application
Data-points Ti = (x(i)1 , x
(i)2 , z(i)), i = 1, . . . ,m
x(i)1 -fat thickness
x(i)2 -lumbar muscle thickness
z (i)- muscle percent of the i th pig
z = a0 + a1x1 + a2x2
outliers
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
Application
Data-points Ti = (x(i)1 , x
(i)2 , z(i)), i = 1, . . . ,m
x(i)1 -fat thickness
x(i)2 -lumbar muscle thickness
z (i)- muscle percent of the i th pig
z = a0 + a1x1 + a2x2
outliers
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
Application
Data-points Ti = (x(i)1 , x
(i)2 , z(i)), i = 1, . . . ,m
x(i)1 -fat thickness
x(i)2 -lumbar muscle thickness
z (i)- muscle percent of the i th pig
z = a0 + a1x1 + a2x2
outliers
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
Root-Mean-Square Error of Prediction (RMSEP)
D. Causeur, G. Daumas, T. Dhorne, B. Engel, M. Font, I. Furnols, S. Hojsgaard, Statistical handbook forassessing pig classification methods: Recommendations from the EUPIGCLASS project group
EU reference method
P. Walstra, G. S. M. Merkus, Procedure for assessment of the lean meat percentage as a consequence of thenew EU reference dissection method in pig carcass classification, DLO-Research Institute for Animal Scienceand Health (ID - DLO). Research Branch Zeist, P.O. Box 501, 3700 AM Zeist, The Netherlands, 1995
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications
LAD ProblemBasic properties
New methodAplication
Example 2 m=145
EU reference method - RMSEP
a∗2 = 0.71259, a∗1 = 0.02312, a∗0 = 67.77137
LAD method
a∗2 = 0.696471, a∗1 = 0.0176915, a∗0 = 67.8906
Kusec, Kuzmanovic, Sabo, Scitovski A new method for searching an L1 solution of an overdetermined system of linear equations and applications