Abstract
We consider a linear regression model with a vector of bounded parameters and a centered noise vector that has an uncertain unimodal distribution but known covariance matrix. We pose the minimax estimation problem for a linear combination of unknown parameters with the use of the probability criterion. The minimax estimate is determined as a result of minimizing a probability bound over the region of possible values of the variance and squared bias for all possible linear estimates. We establish that the resulting robust solution is less conservative in comparison with wider classes of distributions.
Similar content being viewed by others
Change history
10 February 2021
An Erratum to this paper has been published: https://doi.org/10.1134/S0005117920120103
References
Bahadur, R. R., On the Asymptotic Efficiency of Tests and Estimates. Sankhya Indian J. Statist., 1960, Vol. 22, no. 3-4, pp. 229–252.
Ibragimov, I. A. & Khas’minskii, R. Z., Asimptoticheskaya teoriya otsenivaniya Asymptotic Estimation Theory, Moscow: Nauka, 1977).
Vapnik, V. N. & Chervonenkis, A. Ya., Teoriya raspoznavaniya obrazov (statisticheskie problemy obucheniya). (Pattern Recognition Theory: Statistical Learning Problems), Moscow: Nauka, 1974.
Bakhshiyan, B. Ts., Nazirov, R. R., & El’yasberg, P. E., Opredelenie i korrektsiya dvizheniya (Identification and Correction of Motion), Moscow: Nauka, 1980).
Timofeeva, G. A., Nonlinear Confidence Estimates for Statistically Uncertain Systems, Autom. Remote Control, 2003, vol. 64 no. 11, pp. 1724–1733.
Medvedeva, N. V. & Timofeeva, G. A., Comparison of Linear and Nonlinear Methods of Confidence Estimation for Statistically Uncertain Systems, Autom. Remote Control, 2007, vol. 68 no. 4, pp. 619–627.
Anan’ev, B. I., Multistep Specific Stochastic Inclusions and Their Multiestimates, Autom. Remote Control, 2007, vol. 68 no. 11, pp. 1891–1899.
Pankov, A. R. & Semenikhin, K. V., Minimax Estimation by Probabilistic Criterion, Autom. Remote Control, 2007, vol. 68 no. 3, pp. 430–445.
Delage, E. & Ye, Y., Distributionally Robust Optimization under Moment Uncertainty with Application to Data-Driven Problems, Oper. Res., 2010, vol. 58, pp. 595–612.
Kogan, M. M., Robust Estimation and Filtering in Uncertain Linear Systems under Unknown Covariations, Autom. Remote Control, 2015, vol. 76 no. 10, pp. 1751–1764.
Karlin, S. & Studden, W. J., Tchebycheff Systems with Applications in Analysis and Statistics, New York: Interscience, 1966. Translated under the title Chebyshevskie sistemy i ikh primenenie v analize i statistike, Moscow: Nauka, 1976).
Dharmadhikari, S. & Joag-dev, K., Unimodality, Convexity, and Applications, San Diego: Academic, 1988).
Barmish, B. R. & Lagoa, C. M., The Uniform Distribution: A Rigorous Justification for Its Use in Robustness Analysis, Math. Control Signal. Syst., 1997, vol. 10, pp. 203–222.
Kibzun, A. I., On the Worst-Case Distribution in Stochastic Optimization Problems with Probability Function, Autom. Remote Control, 1998, vol. 59 no. 11, pp. 1587–1597.
Kan, Yu. S., On the Justification of the Uniformity Principle in the Optimization of a Probability Performance Index, Autom. Remote Control, 2000, vol. 61 no. 1, pp. 50–64.
Van Parys, B. P. G., Goulart, P. J. & Kuhn, D. Generalized Gauss Inequalities via Semidefinite Programming, Math. Program., 2016, vol. 156, pp. 271–302.
Granichin, O. N., The Nonasymptotic Confidence Set for Parameters of a Linear Control Object under an Arbitrary External Disturbance, Autom. Remote Control, 2012, vol. 73 no. 1, pp. 20–30.
Weyer, E., Campi, M. C., & Csaji, B. C., Asymptotic Properties of SPS Confidence Regions, Automatica, 2017, vol. 82, pp. 287–294.
Semenikhin, K. V., Two-Sided Probability Bound for a Symmetric Unimodal Random Variable, Autom. Remote Control, 2019, vol. 80 no. 3, pp. 474–489.
Vysochanskii, D. F. & Petunin, Yu. I. On a Gauss Inequality for Unimodal Distributions, Theory Probab. Appl., 1983, vol. 27 no. 2, pp. 359–361.
Pukelsheim, F., The Three Sigma Rule, Am. Statist. 1994, vol. 48, pp. 88–91.
Solov’ev, V. N., Dual Extremal Problems and Their Applications to Minimax Estimation Problems, Russ. Math. Surv., 1997, vol. 52 no. 4, pp. 685–720.
Matasov, A. I., Estimators for Uncertain Dynamic Systems, Dordrecht: Kluwer, 1998.
Grant, M.C. Boyd, S.P., The CVX Users’ Guide. Release 2.1, CVX Research, Inc. 2018 [Online]. http://cvxr.com/cvx.
Akimov, P. A. & Matasov, A. I., An Iterative Algorithm for ℓ1-Norm Approximation in Dynamic Estimation Problems, Autom. Remote Control, 2015, vol. 76, no. 5, pp. 733–748.
Arkhipov, A. S. & Semenikhin, K. V., Confidence Analysis of Linear Unbiased Estimates under Uncertain Unimodal Noise Distributions, J. Comput. Syst. Sci. Int., 2019, vol. 58, no. 5, pp. 674–683.
Rockafellar, R. T., Convex Analysis, Princeton: Princeton Univ. Press, 1970. Translated under the title Vypuklyi analiz. Moscow: Mir, 1973.
Author information
Authors and Affiliations
Appendix
Appendix
Proof of Theorem 1. Consider the linear estimate \(\tilde{X}\) defined by the coefficient vector \(f\in {{\mathbb{R}}}^{n}\) and the bias \(c\in {\mathbb{R}}\) according to (5).
If the estimated value X and the observation vector Y satisfy the equations of the regression model (1) with parameter vector θ ∈ Θ and noise vector \(\eta \sim {\mathcal{H}}(K)\), then error \(\tilde{X}-X\) admits the representation ε + r, where \(\varepsilon \sim {\mathcal{H}}(d)\), and d, r satisfy the relations (9). Therefore, Eq. (11) has the inequality sign “⩽.”
To prove the opposite inequality, it suffices for a given vector of parameters θ ∈ Θ and a random variable \(\varepsilon \sim {\mathcal{H}}(d)\), where d and r have the form (9), to choose a random vector \(\eta \sim {\mathcal{H}}(K)\) satisfying equality \(\varepsilon +r=\tilde{X}-X\) with probability 1. By virtue of (1) and (9) the required equality is equivalent to the following:
Acting in the same way as in [26], we define the desired vector by the rule
where P = In − ∣g∣−2gg*, g = K1/2f, In is the identity matrix of size n × n, and ζ is the standard n-dimensional Gaussian random vector independent of ε. Checking conditions ε = ⟨f, η⟩, Mη = 0 and cov{η, η} = K is identical to the calculations from [26].
In case when \({\mathcal{H}}={\mathcal{P}}\), the proof ends here.
With \({\mathcal{H}}={\mathcal{U}}\) it remains to verify that the distribution of the vector η is symmetric and linearly unimodal. According to [12], this condition means that for any choice of the coefficient vector \(b\in {{\mathbb{R}}}^{n}\) the distribution of the linear combination
is symmetric and unimodal. This fact follows from the unimodality of the convolution of two symmetric unimodal one-dimensional distributions that are the distributions of both terms due to \(\varepsilon \sim {\mathcal{U}}(d)\) and \(\zeta \sim {\mathcal{N}}(0,{I}_{n})\) (see Theorem 1.6 from the same source).
Proof of Theorem 2. The convexity of the region Q follows directly from the convexity in \((f,c)\in {{\mathbb{R}}}^{n}\times {\mathbb{R}}\) of two functions
Therefore, the function ρ(d) as the lower boundary of the convex set Q will be convex as well (see Theorem I.5.3 from [27]). And since it is finite everywhere, it will be continuous.
The second statement follows from the definition of a support line for a convex set. With a fixed λ ⩾0, the straight line ρ + λd = γλ is a support line to the set Q at the point (dλ, ρλ) if the linear form ρ + λd reaches a minimum (or maximum) over Q at the specified point. The case of a maximum can be discarded since this linear form on the region Q is not bounded from above. Thus, we obtain the required facts (19) and (20).
Proof of Theorem 3. Theorem 1 due to the monotonic dependence of \({\pi }_{h}^{{\mathcal{H}}}(d,\rho )\) on ρ implies that
where d = ⟨Kf, f⟩ and \(\rho =\mathop{\sup }\limits_{\theta \in \Theta }{\left(c+\langle {B}^{* }f-a,\theta \rangle \right)}^{2}\).
Let \((\hat{d},\hat{\rho })\) be the minimum point of \({\pi }_{h}^{{\mathcal{H}}}(d,\rho )\) over (d, ρ) ∈ Q, and let \((\hat{f},\hat{c})\) be the solution to the minimax problem (19) with parameter λ equal to the slope of the support line for the set Q at point \((\hat{d},\hat{\rho })\). Then, according to Theorem 2, it holds that
Therefore, the estimate \(\hat{X}=\langle \hat{f},Y\rangle +\hat{c}\) realizes the equality
Now, since \({\pi }_{h}^{{\mathcal{H}}}(\hat{d},\hat{\rho })\leqslant {\pi }_{h}^{{\mathcal{H}}}(d,\rho )\), we see that the left-hand side of (A.2) does not exceed the left-hand side of (A.1). Therefore, \(\hat{X}\) is a minimax estimate by the probability criterion on the class \({\mathcal{H}}(K)\), as needed.
Rights and permissions
About this article
Cite this article
Arkhipov, A., Semenikhin, K. Minimax Linear Estimation with the Probability Criterion under Unimodal Noise and Bounded Parameters. Autom Remote Control 81, 1176–1191 (2020). https://doi.org/10.1134/S0005117920070024
Received:
Revised:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1134/S0005117920070024