of 24

• date post

30-Oct-2019
• Category

## Documents

• view

7

0

Embed Size (px)

### Transcript of Complementairty Formulations for - Eaton.math.rpi. Complementairty Formulations for L0-norm...

• Complementairty Formulations for L0-norm Optimization Problems1

John E. Mitchell

Department of Mathematical Sciences RPI, Troy, NY 12180 USA

SIAM Conference on Optimization San Diego, May 2014

1Joint work with Mingbin Feng, Jong-Shi Pang, Xin Shen,and Andreas Wächter. Supported by AFOSR and NSF.

Mitchell (RPI) Complementarity for L0 SIOPT 2014 1 / 22

• Outline

1 Introduction and Applications

2 L0-norm minimization

3 Conclusions

Mitchell (RPI) Complementarity for L0 SIOPT 2014 2 / 22

• Introduction and Applications

Outline

1 Introduction and Applications

2 L0-norm minimization

3 Conclusions

Mitchell (RPI) Complementarity for L0 SIOPT 2014 3 / 22

• Introduction and Applications

L0-norm minimization

Want to solve:

min x {f (x) + γ||x ||0 : Ax ≥ b, x ∈ IRn}

where A ∈ IRm×n, b ∈ IRm, and ||x ||0 = card{xi : xi 6= 0}.

The parameter γ reflects the relative importance of sparsity and f (x).

L0-norm minimization is of interest in feature selection, compressed sensing, misclassification minimization, and almost any problem where a sparse solution is desired.

Can be modeled as a Mathematical Program with Complementarity Constraints or MPCC.

Mitchell (RPI) Complementarity for L0 SIOPT 2014 4 / 22

• Introduction and Applications

Eg: Misclassification minimization

Given (C, γ, ε) > 0, and points x i with observed values yi :

minimize (w ,b)

C n∑

i=1

max { |wT x i + b − yi | − ε, 0

} + 12 w

T w︸ ︷︷ ︸ standard SVM objective

+ γ n∑

i=1

{ 1 if |wT x i + b − yi | > ε

0 if |wT x i + b − yi | ≤ ε︸ ︷︷ ︸ number of misclassified points

(Mangasarian)

Mitchell (RPI) Complementarity for L0 SIOPT 2014 5 / 22

• Introduction and Applications

Eg: Minimum portfolio revision

Given portfolio x0 in standard unit simplex ∆n , {x ∈ Rn+ | 1T x = 1}, consider a portfolio revision problem as follows:

minimize x∈∆n

c 2

xT Vx︸ ︷︷ ︸ variance

+c ′ VaRβ(x)

︸ ︷︷ ︸ composite risk measure

+ K ‖ x − x0 ‖0︸ ︷︷ ︸ portfolio revision cost

subject to µT x ≥ R (expected portfolio return),

where VaRβ(x) is the β-value-at-risk of the portfolio x with β ∈ (0,1).

Mitchell (RPI) Complementarity for L0 SIOPT 2014 6 / 22

• L0-norm minimization

Outline

1 Introduction and Applications

2 L0-norm minimization

3 Conclusions

Mitchell (RPI) Complementarity for L0 SIOPT 2014 7 / 22

• L0-norm minimization Formulation

Complementarity formulation Want to solve:

min x {f (x) + γ||x ||0 : Ax ≥ b, x ∈ IRn}

Equivalently:

minx±,ξ f (x) + ∑n

i=1(1− ξi) subject to Ax ≥ b

0 ≤ ξ ≤ 1 0 ≤ ξ ⊥ x

Note that if xi = 0 then we can choose ξi = 1, and if xi 6= 0 then we must set ξi = 0.

Thus, the objective counts the number of nonzero components.

No need for any big-M terms in this LPCC formulation. Mitchell (RPI) Complementarity for L0 SIOPT 2014 8 / 22

• L0-norm minimization Formulation

Complementarity formulation

minx±,ξ f (x) + ∑n

i=1(1− ξi) subject to Ax ≥ b

0 ≤ ξ ≤ 1 0 ≤ ξ ⊥ x

This is a half-complementary formulation;

Mitchell (RPI) Complementarity for L0 SIOPT 2014 9 / 22

• L0-norm minimization Formulation

Complementarity formulation

minx±,ξ f (x) + ∑n

i=1(1− ξi) subject to Ax ≥ b

0 ≤ ξ ≤ 1 0 ≤ ξ ⊥ x+ + x− ≥ 0, 0 ≤ x+ ⊥ x− ≥ 0, x = x+ − x−

This is a half-complementary formulation; can also get a full-complementary formulation by splitting x = x+ − x−.

Mitchell (RPI) Complementarity for L0 SIOPT 2014 9 / 22

• L0-norm minimization Formulation

KKT points

The Constant Rank Constraint Qualification holds. It also holds under additional assumptions with nonlinear constraints.

Let x satisfy Ax ≥ b and assume x minimizes f (x) for some assignment of the complementarities, set ξ to count the number of nonzero components:

This point is a local minimizer and a strongly stationary point of the MPCC formulation.

Mitchell (RPI) Complementarity for L0 SIOPT 2014 10 / 22

• L0-norm minimization Formulation

Relaxed NLP formulation Let � > 0. Assume f (x) ≡ 0. Can get a relaxed formulation:

minx±,ξ ∑n

i=1(1− ξi) subject to Ax+ − Ax− ≥ b

0 ≤ x+, x−

0 ≤ ξ ≤ 1 ξi (x+i + x

− i ) ≤ � i = 1, . . . ,n

CQ holds for this problem.

Not every x is a local minimizer.

A sequence of global minimizers as �→ 0 will have a subsequence that converges to a global minimizer of the L0-norm problem.

Under certain conditions, the solution to the relaxed formulation is a solution to a reweighted version of the L1-norm minimization problem.

Mitchell (RPI) Complementarity for L0 SIOPT 2014 11 / 22

• L0-norm minimization Formulation

Penalty method

Can penalize violations of complementarity in many ways:

• Add (x+ + x−)T ξ to the objective function. • Add (x+ + x−)T ξ + (x+)T x− to the objective function.

Corresponds to the AMPL keyword “complements” when using knitro.

• Add (((x+ + x−)T ξ)2 to the objective function. • Add ((x+ + x−)T ξ + (x+)T x−))2 to the objective function. • Use the Fischer-Burmeister function, or some other NCP function.

Theoretical investigation is ongoing.

Surprisingly, the fourth formulation has worked well in our tests.

Mitchell (RPI) Complementarity for L0 SIOPT 2014 12 / 22

• L0-norm minimization Formulation

Penalty method

Can penalize violations of complementarity in many ways:

• Add (x+ + x−)T ξ to the objective function. • Add (x+ + x−)T ξ + (x+)T x− to the objective function.

Corresponds to the AMPL keyword “complements” when using knitro.

• Add (((x+ + x−)T ξ)2 to the objective function. • Add ((x+ + x−)T ξ + (x+)T x−))2 to the objective function. • Use the Fischer-Burmeister function, or some other NCP function.

Theoretical investigation is ongoing.

Surprisingly, the fourth formulation has worked well in our tests.

Mitchell (RPI) Complementarity for L0 SIOPT 2014 12 / 22

• L0-norm minimization Computational Results

Test problems

Examine three classes of problems:

• Minimize L0-norm, without another objective f (x).

• Look at weighted combinations of f (x) and L0-norm.

• Look at signal recovery problems.

In each case, compare solutions obtained by a nonlinear programming package (KNITRO, SNOPT, CONOPT, MINOS) for our complementarity formulation with an L1-norm formulation.

Have no guarantee of global optimality.

Mitchell (RPI) Complementarity for L0 SIOPT 2014 13 / 22

• L0-norm minimization Minimize L0-norm

Minimize number of non zeroes

minx∈IRn ‖x‖0,1 s.t. Ax ≥ b.

Look at four formulations:

MILP: integer programming formulation with a big-M

L1: L1-approximation solved as an LP. Objective reweighted iteratively (Candes et al).

fullcomp: complementarity formulation solved as NLP

halfcomp: modified complementarity formulation, solved as NLP

Test problems A and b generated randomly, each entry in U[−1,1].

Mitchell (RPI) Complementarity for L0 SIOPT 2014 14 / 22

• L0-norm minimization Minimize L0-norm

A is 30× 50: 50 test problems, various solvers and start points

Mitchell (RPI) Complementarity for L0 SIOPT 2014 15 / 22

• L0-norm minimization Minimize L0-norm

A is 300× 500: performance profile of sparsity

Mitchell (RPI) Complementarity for L0 SIOPT 2014 16 / 22

• L0-norm minimization Pareto frontier

Weighted combinations

minx∈IRn q(x) + β‖x‖0,1 s.t. Ax ≥ b.

Solve for each norm for various different choices of β.

In our tests, A is 30× 50, q(x) is strictly convex quadratic function.

The L0-norm formulations are solved using KNITRO under AMPL.

Mitchell (RPI) Complementarity for L0 SIOPT 2014 17 / 22

• L0-norm minimization Pareto frontier

Pareto frontier

Mitchell (RPI) Complementarity for L0 SIOPT 2014 18 / 22

• L0-norm minimization Signal recovery

Signal recovery

min x∈IRn

‖Ax − b‖22 + β‖x‖0,1

Look at different choices of regularizing parameter β.

A is 256× 1024. In original signal, 40 entries of x are nonzero. Random noise is added to give b.

L1-norm results are improved by debiasing: fix zeroes in L1 solution at zero, then look for best least squares fit. L1-debiased leads to many small nonzero components in our tests.

Mitchell (RPI) Complementarity for L0 SIOPT 2014 19 / 22

• L0-norm minimization Signal recovery

Signal recovery sparsity

10−6 10−4 10−2 100 100

101

102

103

104

y=40

Regularizing Paramet