American Institute of Mathematical Sciences

• Previous Article
Using generalized cross validation to select regularization parameter for total variation regularization problems
• IPI Home
• This Issue
• Next Article
Recovery of seismic wavefields by an lq-norm constrained regularization method
October  2018, 12(5): 1121-1155. doi: 10.3934/ipi.2018047

Risk estimators for choosing regularization parameters in ill-posed problems - properties and limitations

 1 Centrum Wiskunde & Informatica (CWI), Science Park 123, 1098 XG Amsterdam, The Netherlands 2 Department of Computer Science, University College London, WC1E 6BT London, UK 3 Universität Göttingen, Institut für Mathematische Stochastik, Goldschmidtstrasse 7, 37077 Göttingen, Germany 4 Department of Applied Mathematics, University of Twente, P.O. Box 217, 7500 AE Enschede, The Netherlands 5 Fakultät für Mathematik, Ruhr-Universität, Bochum, 44780 Bochum, Germany 6 Institut für Numerische und Angewandte Mathematik, Westfälische Wilhelms-Universität (WWU) Münster, Einsteinstr. 62, D 48149 Münster, Germany

* Corresponding author: felix.lucka@cwi.nl

Received  January 2017 Revised  May 2018 Published  July 2018

This paper discusses the properties of certain risk estimators that recently regained popularity for choosing regularization parameters in ill-posed problems, in particular for sparsity regularization. They apply Stein's unbiased risk estimator (SURE) to estimate the risk in either the space of the unknown variables or in the data space. We will call the latter PSURE in order to distinguish the two different risk functions. It seems intuitive that SURE is more appropriate for ill-posed problems, since the properties in the data space do not tell much about the quality of the reconstruction. We provide theoretical studies of both approaches for linear Tikhonov regularization in a finite dimensional setting and estimate the quality of the risk estimators, which also leads to asymptotic convergence results as the dimension of the problem tends to infinity. Unlike previous works which studied single realizations of image processing problems with a very low degree of ill-posedness, we are interested in the statistical behaviour of the risk estimators for increasing ill-posedness. Interestingly, our theoretical results indicate that the quality of the SURE risk can deteriorate asymptotically for ill-posed problems, which is confirmed by an extensive numerical study. The latter shows that in many cases the SURE estimator leads to extremely small regularization parameters, which obviously cannot stabilize the reconstruction. Similar but less severe issues with respect to robustness also appear for the PSURE estimator, which in comparison to the rather conservative discrepancy principle leads to the conclusion that regularization parameter choice based on unbiased risk estimation is not a reliable procedure for ill-posed problems. A similar numerical study for sparsity regularization demonstrates that the same issue appears in non-linear variational regularization approaches.

Citation: Felix Lucka, Katharina Proksch, Christoph Brune, Nicolai Bissantz, Martin Burger, Holger Dette, Frank Wübbeling. Risk estimators for choosing regularization parameters in ill-posed problems - properties and limitations. Inverse Problems & Imaging, 2018, 12 (5) : 1121-1155. doi: 10.3934/ipi.2018047
References:
 [1] R. J. Adler and J. E. Taylor, Random Fields and Geometry, Springer Monographs in Mathematics, Springer, New York, 2007. Google Scholar [2] M. S. C. Almeida and M. A. T. Figueiredo, Parameter estimation for blind and non-blind deblurring using residual whiteness measures, IEEE Transactions on Image Processing, 22 (2013), 2751-2763. doi: 10.1109/TIP.2013.2257810. Google Scholar [3] F. Bauer and T. Hohage, A Lepskij-type stopping rule for regularized Newton methods, Inverse Problems, 21 (2005), 1975-1991. doi: 10.1088/0266-5611/21/6/011. Google Scholar [4] G. Blanchard and P. Mathé, Discrepancy principle for statistical inverse problems with application to conjugate gradient iteration Inverse Problems, 28 (2012), 115011, 23pp. doi: 10.1088/0266-5611/28/11/115011. Google Scholar [5] P. Blomgren and T. F. Chan, Modular solvers for image restoration problems using the discrepancy principle, Numerical Linear Algebra with Applications, 9 (2002), 347-358. doi: 10.1002/nla.278. Google Scholar [6] S. Boyd, N. Parikh, E. Chu, B. Peleato and J. Eckstein, Distributed optimization and statistical learning via the alternating direction method of multipliers, Foundations and Trends in Machine Learning, 3 (2011), 1-122. doi: 10.1561/2200000016. Google Scholar [7] B. Bringmann, D. Cremers, F. Krahmer and M. Möller, The homotopy method revisited: Computing solution paths of $\ell_1$-regularized problems, Math. Comp., 87 (2018), 2343-2364, arXiv: 1605.00071. doi: 10.1090/mcom/3287. Google Scholar [8] M. Burger, A. Sawatzky and G. Steidl, First order algorithms in variational image processing, Splitting Methods in Communication, Imaging, Science, and Engineering, 345-407, Sci. Comput., Springer, Cham, 2016. Google Scholar [9] E. J. Candes, C. A. Sing-Long and J. D. Trzasko, Unbiased risk estimates for singular value thresholding and spectral estimators, IEEE Transactions on Signal Processing, 61 (2013), 4643-4657. doi: 10.1109/TSP.2013.2270464. Google Scholar [10] E. Chernousova and Y. Golubev, Spectral cut-off regularizations for ill-posed linear models, Math. Methods Statist., 23 (2014), 116-131. doi: 10.3103/S1066530714020033. Google Scholar [11] C. Deledalle, S. Vaiter, J. Fadili and G. Peyré, Stein Unbiased GrAdient estimator of the Risk (SUGAR) for Multiple Parameter Selection, SIAM Journal on Imaging Sciences, 7 (2014), 2448-2487. doi: 10.1137/140968045. Google Scholar [12] C. Deledalle, S. Vaiter, G. Peyré, J. Fadili and C. Dossal, Proximal splitting derivatives for risk estimation, Journal of Physics: Conference Series, 386 (2012), 012003. doi: 10.1088/1742-6596/386/1/012003. Google Scholar [13] C. Deledalle, S. Vaiter, G. Peyré, J. Fadili and C. Dossal, Unbiased risk estimation for sparse analysis regularization, in 2012 19th IEEE International Conference on Image Processing, IEEE, 2012, 3053-3056. doi: 10.1109/ICIP.2012.6467544. Google Scholar [14] C. Dossal, M. Kachour, J. Fadili, G. Peyré and C. Chesneau, The degrees of freedom of the lasso for general design matrix, Statistica Sinica, 23 (2013), 809-828. Google Scholar [15] Y. C. Eldar, Generalized SURE for Exponential Families: Applications to Regularization, IEEE Transactions on Signal Processing, 57 (2009), 471-481. doi: 10.1109/TSP.2008.2008212. Google Scholar [16] H. W. Engl, M. Hanke and A. Neubauer, Regularization of Inverse Problems, vol. 375, Springer Science & Business Media, 1996.Google Scholar [17] N. P. Galatsanos and A. K. Katsaggelos, Methods for choosing the regularization parameter and estimating the noise variance in image restoration and their relation, Trans. Img. Proc., 1 (1992), 322-336. doi: 10.1109/83.148606. Google Scholar [18] S. K. Ghoreishi and M. R. Meshkani, On SURE estimates in hierarchical models assuming heteroscedasticity for both levels of a two-level normal hierarchical model, Journal of Multivariate Analysis, 132 (2014), 129-137. doi: 10.1016/j.jmva.2014.08.001. Google Scholar [19] R. Giryes, M. Elad and Y. Eldar, The projected GSURE for automatic parameter tuning in iterative shrinkage methods, Applied and Computational Harmonic Analysis, 30 (2011), 407-422. doi: 10.1016/j.acha.2010.11.005. Google Scholar [20] J. Hadamard, Lectures on Cauchy's Problem in Linear Partial Differential Equations, New Haven, 1953. Google Scholar [21] H. Haghshenas Lari and A. Gholami, Curvelet-TV regularized Bregman iteration for seismic random noise attenuation, Journal of Applied Geophysics, 109 (2014), 233-241. doi: 10.1016/j.jappgeo.2014.08.005. Google Scholar [22] P. C. Hansen, Analysis of discrete ill-posed problems by means of the L-curve, SIAM Review, 34 (1992), 561-580. doi: 10.1137/1034115. Google Scholar [23] P. C. Hansen and D. P. OLeary, The Use of the L-Curve in the Regularization of Discrete Ill-Posed Problems, SIAM Journal on Scientific Computing, 14 (1993), 1487-1503. doi: 10.1137/0914086. Google Scholar [24] B. Jin, J. Zou et al., Iterative parameter choice by discrepancy principle, IMA Journal of Numerical Analysis, 32 (2012), 1714-1732. doi: 10.1093/imanum/drr051. Google Scholar [25] A. Kneip, Ordered linear smoothers, Ann. Statist., 22 (1994), 835-866. doi: 10.1214/aos/1176325498. Google Scholar [26] O. V. Lepskii, On a Problem of Adaptive Estimation in Gaussian White Noise, Theory of Probability & Its Applications, 35 (1991), 454-466. doi: 10.1137/1135065. Google Scholar [27] H. Li and F. Werner, Empirical risk minimization as parameter choice rule for general linear regularization methods, 2017, arXiv: 1703.07809.Google Scholar [28] K.-C. Li, From stein's unbiased risk estimates to the method of generalized cross validation, The Annals of Statistics, 13 (1985), 1352-1377. doi: 10.1214/aos/1176349742. Google Scholar [29] K.-C. Li, Asymptotic optimality for $C_p$, $C_L$, cross-validation and generalized cross-validation: Discrete index set, Ann. Statist., 15 (1987), 958-975. doi: 10.1214/aos/1176350486. Google Scholar [30] F. Luisier, T. Blu and M. Unser, Image denoising in mixed Poisson-Gaussian noise, IEEE Transactions on Image Processing, 20 (2011), 696-708. doi: 10.1109/TIP.2010.2073477. Google Scholar [31] J.-C. Pesquet, A. Benazza-Benyahia and C. Chaux, A SURE Approach for Digital Signal/Image Deconvolution Problems, IEEE Transactions on Signal Processing, 57 (2009), 4616-4632. doi: 10.1109/TSP.2009.2026077. Google Scholar [32] P. Qu, C. Wang and G. X. Shen, Discrepancy-based adaptive regularization for grappa reconstruction, Journal of Magnetic Resonance Imaging, 24 (2006), 248-255. doi: 10.1002/jmri.20620. Google Scholar [33] S. Ramani, T. Blu and M. Unser, Monte-Carlo sure: A black-box optimization of regularization parameters for general denoising algorithms, IEEE Transactions on Image Processing, 17 (2008), 1540-1554. doi: 10.1109/TIP.2008.2001404. Google Scholar [34] S. Ramani, Z. Liu, J. Rosen, J.-F. Nielsen and J. A. Fessler, Regularization parameter selection for nonlinear iterative image restoration and MRI reconstruction using GCV and SURE-based methods, IEEE Transactions on Image Processing, 21 (2012), 3659-3672. doi: 10.1109/TIP.2012.2195015. Google Scholar [35] J. A. Rice, Choice of smoothing parameter in deconvolution problems, Contemporary Mathematics, 59 (1986), 137-151. doi: 10.1090/conm/059/10. Google Scholar [36] C. M. Stein, Estimation of the mean of a multivariate normal distribution, The Annals of Statistics, 9 (1981), 1135-1151. doi: 10.1214/aos/1176345632. Google Scholar [37] A. M. Thompson, J. C. Brown, J. W. Kay and D. M. Titterington, A study of methods of choosing the smoothing parameter in image restoration by regularization, IEEE Trans. Pattern Anal. Mach. Intell., 13 (1991), 326-339. doi: 10.1109/34.88568. Google Scholar [38] G. M. Vainikko, The discrepancy principle for a class of regularization methods, USSR Computational Mathematics and Mathematical Physics, 22 (1982), 1-19. doi: 10.1016/0041-5553(82)90120-3. Google Scholar [39] S. Vaiter, C. Deledalle and G. Peyré, The degrees of freedom of partly smooth regularizers, Annals of the Institute of Statistical Mathematics, 69 (2017), 791-832. doi: 10.1007/s10463-016-0563-z. Google Scholar [40] S. Vaiter, C. Deledalle, G. Peyré, C. Dossal and J. Fadili, Local behavior of sparse analysis regularization: Applications to risk estimation, Applied and Computational Harmonic Analysis, 35 (2013), 433-451. doi: 10.1016/j.acha.2012.11.006. Google Scholar [41] S. A. vande Geer, Applications of Empirical Process Theory, vol. 6 of Cambridge Series in Statistical and Probabilistic Mathematics, Cambridge University Press, Cambridge, 2000. Google Scholar [42] D. Van De Ville and M. Kocher, SURE-Based Non-Local Means, IEEE Signal Processing Letters, 16 (2009), 973-976, URL http://ieeexplore.ieee.org/lpdocs/epic03/wrapper.htm?arnumber=5165022.Google Scholar [43] D. Van DeVille and M. Kocher, Nonlocal means with dimensionality reduction and SURE-based parameter selection, IEEE Transactions on Image Processing, 20 (2011), 2683-2690. doi: 10.1109/TIP.2011.2121083. Google Scholar [44] Y.-Q. Wang and J.-M. Morel, SURE Guided Gaussian Mixture Image Denoising, SIAM Journal on Imaging Sciences, 6 (2013), 999-1034. doi: 10.1137/120901131. Google Scholar [45] D. S. Weller, S. Ramani, J.-F. Nielsen and J. A. Fessler, Monte Carlo SURE-based parameter selection for parallel magnetic resonance imaging reconstruction, Magnetic Resonance in Medicine, 71 (2014), 1760-1770. Google Scholar [46] X. Xie, S. C. Kou and L. D. Brown, SURE Estimates for a Heteroscedastic Hierarchical Model, Journal of the American Statistical Association, 107 (2012), 1465-1479. doi: 10.1080/01621459.2012.728154. Google Scholar

show all references

References:
 [1] R. J. Adler and J. E. Taylor, Random Fields and Geometry, Springer Monographs in Mathematics, Springer, New York, 2007. Google Scholar [2] M. S. C. Almeida and M. A. T. Figueiredo, Parameter estimation for blind and non-blind deblurring using residual whiteness measures, IEEE Transactions on Image Processing, 22 (2013), 2751-2763. doi: 10.1109/TIP.2013.2257810. Google Scholar [3] F. Bauer and T. Hohage, A Lepskij-type stopping rule for regularized Newton methods, Inverse Problems, 21 (2005), 1975-1991. doi: 10.1088/0266-5611/21/6/011. Google Scholar [4] G. Blanchard and P. Mathé, Discrepancy principle for statistical inverse problems with application to conjugate gradient iteration Inverse Problems, 28 (2012), 115011, 23pp. doi: 10.1088/0266-5611/28/11/115011. Google Scholar [5] P. Blomgren and T. F. Chan, Modular solvers for image restoration problems using the discrepancy principle, Numerical Linear Algebra with Applications, 9 (2002), 347-358. doi: 10.1002/nla.278. Google Scholar [6] S. Boyd, N. Parikh, E. Chu, B. Peleato and J. Eckstein, Distributed optimization and statistical learning via the alternating direction method of multipliers, Foundations and Trends in Machine Learning, 3 (2011), 1-122. doi: 10.1561/2200000016. Google Scholar [7] B. Bringmann, D. Cremers, F. Krahmer and M. Möller, The homotopy method revisited: Computing solution paths of $\ell_1$-regularized problems, Math. Comp., 87 (2018), 2343-2364, arXiv: 1605.00071. doi: 10.1090/mcom/3287. Google Scholar [8] M. Burger, A. Sawatzky and G. Steidl, First order algorithms in variational image processing, Splitting Methods in Communication, Imaging, Science, and Engineering, 345-407, Sci. Comput., Springer, Cham, 2016. Google Scholar [9] E. J. Candes, C. A. Sing-Long and J. D. Trzasko, Unbiased risk estimates for singular value thresholding and spectral estimators, IEEE Transactions on Signal Processing, 61 (2013), 4643-4657. doi: 10.1109/TSP.2013.2270464. Google Scholar [10] E. Chernousova and Y. Golubev, Spectral cut-off regularizations for ill-posed linear models, Math. Methods Statist., 23 (2014), 116-131. doi: 10.3103/S1066530714020033. Google Scholar [11] C. Deledalle, S. Vaiter, J. Fadili and G. Peyré, Stein Unbiased GrAdient estimator of the Risk (SUGAR) for Multiple Parameter Selection, SIAM Journal on Imaging Sciences, 7 (2014), 2448-2487. doi: 10.1137/140968045. Google Scholar [12] C. Deledalle, S. Vaiter, G. Peyré, J. Fadili and C. Dossal, Proximal splitting derivatives for risk estimation, Journal of Physics: Conference Series, 386 (2012), 012003. doi: 10.1088/1742-6596/386/1/012003. Google Scholar [13] C. Deledalle, S. Vaiter, G. Peyré, J. Fadili and C. Dossal, Unbiased risk estimation for sparse analysis regularization, in 2012 19th IEEE International Conference on Image Processing, IEEE, 2012, 3053-3056. doi: 10.1109/ICIP.2012.6467544. Google Scholar [14] C. Dossal, M. Kachour, J. Fadili, G. Peyré and C. Chesneau, The degrees of freedom of the lasso for general design matrix, Statistica Sinica, 23 (2013), 809-828. Google Scholar [15] Y. C. Eldar, Generalized SURE for Exponential Families: Applications to Regularization, IEEE Transactions on Signal Processing, 57 (2009), 471-481. doi: 10.1109/TSP.2008.2008212. Google Scholar [16] H. W. Engl, M. Hanke and A. Neubauer, Regularization of Inverse Problems, vol. 375, Springer Science & Business Media, 1996.Google Scholar [17] N. P. Galatsanos and A. K. Katsaggelos, Methods for choosing the regularization parameter and estimating the noise variance in image restoration and their relation, Trans. Img. Proc., 1 (1992), 322-336. doi: 10.1109/83.148606. Google Scholar [18] S. K. Ghoreishi and M. R. Meshkani, On SURE estimates in hierarchical models assuming heteroscedasticity for both levels of a two-level normal hierarchical model, Journal of Multivariate Analysis, 132 (2014), 129-137. doi: 10.1016/j.jmva.2014.08.001. Google Scholar [19] R. Giryes, M. Elad and Y. Eldar, The projected GSURE for automatic parameter tuning in iterative shrinkage methods, Applied and Computational Harmonic Analysis, 30 (2011), 407-422. doi: 10.1016/j.acha.2010.11.005. Google Scholar [20] J. Hadamard, Lectures on Cauchy's Problem in Linear Partial Differential Equations, New Haven, 1953. Google Scholar [21] H. Haghshenas Lari and A. Gholami, Curvelet-TV regularized Bregman iteration for seismic random noise attenuation, Journal of Applied Geophysics, 109 (2014), 233-241. doi: 10.1016/j.jappgeo.2014.08.005. Google Scholar [22] P. C. Hansen, Analysis of discrete ill-posed problems by means of the L-curve, SIAM Review, 34 (1992), 561-580. doi: 10.1137/1034115. Google Scholar [23] P. C. Hansen and D. P. OLeary, The Use of the L-Curve in the Regularization of Discrete Ill-Posed Problems, SIAM Journal on Scientific Computing, 14 (1993), 1487-1503. doi: 10.1137/0914086. Google Scholar [24] B. Jin, J. Zou et al., Iterative parameter choice by discrepancy principle, IMA Journal of Numerical Analysis, 32 (2012), 1714-1732. doi: 10.1093/imanum/drr051. Google Scholar [25] A. Kneip, Ordered linear smoothers, Ann. Statist., 22 (1994), 835-866. doi: 10.1214/aos/1176325498. Google Scholar [26] O. V. Lepskii, On a Problem of Adaptive Estimation in Gaussian White Noise, Theory of Probability & Its Applications, 35 (1991), 454-466. doi: 10.1137/1135065. Google Scholar [27] H. Li and F. Werner, Empirical risk minimization as parameter choice rule for general linear regularization methods, 2017, arXiv: 1703.07809.Google Scholar [28] K.-C. Li, From stein's unbiased risk estimates to the method of generalized cross validation, The Annals of Statistics, 13 (1985), 1352-1377. doi: 10.1214/aos/1176349742. Google Scholar [29] K.-C. Li, Asymptotic optimality for $C_p$, $C_L$, cross-validation and generalized cross-validation: Discrete index set, Ann. Statist., 15 (1987), 958-975. doi: 10.1214/aos/1176350486. Google Scholar [30] F. Luisier, T. Blu and M. Unser, Image denoising in mixed Poisson-Gaussian noise, IEEE Transactions on Image Processing, 20 (2011), 696-708. doi: 10.1109/TIP.2010.2073477. Google Scholar [31] J.-C. Pesquet, A. Benazza-Benyahia and C. Chaux, A SURE Approach for Digital Signal/Image Deconvolution Problems, IEEE Transactions on Signal Processing, 57 (2009), 4616-4632. doi: 10.1109/TSP.2009.2026077. Google Scholar [32] P. Qu, C. Wang and G. X. Shen, Discrepancy-based adaptive regularization for grappa reconstruction, Journal of Magnetic Resonance Imaging, 24 (2006), 248-255. doi: 10.1002/jmri.20620. Google Scholar [33] S. Ramani, T. Blu and M. Unser, Monte-Carlo sure: A black-box optimization of regularization parameters for general denoising algorithms, IEEE Transactions on Image Processing, 17 (2008), 1540-1554. doi: 10.1109/TIP.2008.2001404. Google Scholar [34] S. Ramani, Z. Liu, J. Rosen, J.-F. Nielsen and J. A. Fessler, Regularization parameter selection for nonlinear iterative image restoration and MRI reconstruction using GCV and SURE-based methods, IEEE Transactions on Image Processing, 21 (2012), 3659-3672. doi: 10.1109/TIP.2012.2195015. Google Scholar [35] J. A. Rice, Choice of smoothing parameter in deconvolution problems, Contemporary Mathematics, 59 (1986), 137-151. doi: 10.1090/conm/059/10. Google Scholar [36] C. M. Stein, Estimation of the mean of a multivariate normal distribution, The Annals of Statistics, 9 (1981), 1135-1151. doi: 10.1214/aos/1176345632. Google Scholar [37] A. M. Thompson, J. C. Brown, J. W. Kay and D. M. Titterington, A study of methods of choosing the smoothing parameter in image restoration by regularization, IEEE Trans. Pattern Anal. Mach. Intell., 13 (1991), 326-339. doi: 10.1109/34.88568. Google Scholar [38] G. M. Vainikko, The discrepancy principle for a class of regularization methods, USSR Computational Mathematics and Mathematical Physics, 22 (1982), 1-19. doi: 10.1016/0041-5553(82)90120-3. Google Scholar [39] S. Vaiter, C. Deledalle and G. Peyré, The degrees of freedom of partly smooth regularizers, Annals of the Institute of Statistical Mathematics, 69 (2017), 791-832. doi: 10.1007/s10463-016-0563-z. Google Scholar [40] S. Vaiter, C. Deledalle, G. Peyré, C. Dossal and J. Fadili, Local behavior of sparse analysis regularization: Applications to risk estimation, Applied and Computational Harmonic Analysis, 35 (2013), 433-451. doi: 10.1016/j.acha.2012.11.006. Google Scholar [41] S. A. vande Geer, Applications of Empirical Process Theory, vol. 6 of Cambridge Series in Statistical and Probabilistic Mathematics, Cambridge University Press, Cambridge, 2000. Google Scholar [42] D. Van De Ville and M. Kocher, SURE-Based Non-Local Means, IEEE Signal Processing Letters, 16 (2009), 973-976, URL http://ieeexplore.ieee.org/lpdocs/epic03/wrapper.htm?arnumber=5165022.Google Scholar [43] D. Van DeVille and M. Kocher, Nonlocal means with dimensionality reduction and SURE-based parameter selection, IEEE Transactions on Image Processing, 20 (2011), 2683-2690. doi: 10.1109/TIP.2011.2121083. Google Scholar [44] Y.-Q. Wang and J.-M. Morel, SURE Guided Gaussian Mixture Image Denoising, SIAM Journal on Imaging Sciences, 6 (2013), 999-1034. doi: 10.1137/120901131. Google Scholar [45] D. S. Weller, S. Ramani, J.-F. Nielsen and J. A. Fessler, Monte Carlo SURE-based parameter selection for parallel magnetic resonance imaging reconstruction, Magnetic Resonance in Medicine, 71 (2014), 1760-1770. Google Scholar [46] X. Xie, S. C. Kou and L. D. Brown, SURE Estimates for a Heteroscedastic Hierarchical Model, Journal of the American Statistical Association, 107 (2012), 1465-1479. doi: 10.1080/01621459.2012.728154. Google Scholar
(a) The convolution kernel $k_l(t)$ for different values of $l$.(b) True solution $x^*$, clean data $A_l x^*$ and noisy data $A_lx^* +\varepsilon$ for $m = n = 64$, $l = 0.06$, $\sigma = 0.1$
Decay of the singular values $\gamma_i$ of $A_l$ for different choices of $m = n$ and $l$. As expected, increasing the width $l$ of the convolution kernel leads to a faster decay. For a fixed $l$, increasing $m$ corresponds to using a finer discretization and $\gamma_i$ converges to the corresponding singular value of $A_{\infty, l}$, as can be seen for the largest $\gamma_i$, e.g., for $l = 0.02$
Empirical probabilities of (a) $\hat \alpha$ and (b) the corresponding $\ell_2$-error for different parameter choice rules using $m = n = 64$, $l = 0.06$, $\sigma = 0.1$ and $N_\varepsilon = 10^6$ samples of $\varepsilon$
Joint empirical probabilities of $\log_{10}\|x^* - x_{\hat \alpha}\|_2$ using $m = n = 64$, $l = 0.06$, $\sigma = 0.1$ and $N_\varepsilon = 10^6$ samples of $\varepsilon$ (the histograms in Figure 3(b) are the marginal distributions thereof). As in Figure 3(b), the logarithms of the probabilities are displayed (here in form of a color-coding) to facilitate the identification of smaller modes and tails. The red line at $x = y$ divides the areas where one method performs better than the other: In (a), all samples falling into the area on the right of the red line correspond to a noise realization where the discrepancy principle leads to a smaller error than PSURE. The percentage of samples for which that is true is 13% for PSURE and 44% for SURE
True risk functions (black dotted line), their estimates for six different realizations $y^k$, $k = 1\ldots 6$ (solid lines), and their corresponding minima/roots (dots on the lines) in the setting described in Figure 1 using $\ell_2$-regularization: (a) $\text{DP}(\alpha, Ax^*)$ and $\text{DP}(\alpha, y^k)$. (b) $\text{MSPE}(\alpha)$ and $\text{PSURE}(\alpha, y^k)$. (c) $\text{MSEE}(\alpha)$ and $\text{SURE}(\alpha, y^k)$
Illustration of Theorems 3.3 and 3.8 for $\ell_2$-regularization: The left hand side of (22)/(23) was estimated by the sample mean and plotted vs. $m$. For (23), the normalization with $\text{cond}(A)$ was omitted in (b) and included in (c). The black dotted lines were added to compare the order of convergence
Empirical probabilities of $\alpha$ for $\ell_2$-regularization and different parameter choice rules for $l = 0.06$ and varying $m$
Empirical probabilities of $\log_{10}\left(\tfrac{1}{m} \| x^* - x_{\alpha} \|_2^2\right)$ for $\ell_2$-regularization and different parameter choice rules for $l = 0.06$ and varying $m$
Empirical probabilities of $\alpha$ for $\ell_2$-regularization and different parameter choice rules for $m = 64$ and varying $l$
Empirical probabilities of $\log_{10}\left(\tfrac{1}{m} \| x^* - x_{\alpha} \|_2^2\right)$ for $\ell_2$-regularization and different parameter choice rules for $m = 64$ and varying $l$
Illustration of the difference between evaluating the SURE risk on a coarse, linear grid for $\alpha$ as opposed to a fine, logarithmic one: In (a), a linear grid is constructed around ${{\hat{\alpha }}_{\text{DP}}}$ as $\alpha = \Delta_\alpha, 2 \Delta_\alpha, \ldots, 50\Delta_\alpha$ with $\Delta_\alpha = 2 {{\hat{\alpha }}_{\text{DP}}}/50$. While the plot suggests a clear minimum, (b) reveals that it is only a sub-optimal local minimum and that the linear grid did not cover the essential parts of $\text{SURE}(\alpha, y)$. (c) and (d) show the same plots for a different noise realization. Here, a linear grid will not even find a clear minimum. Both risk estimators are the same as those plotted in Figure 5(c) with the same colors
Risk functions (black dotted line), $k = 1, \ldots, 6$ estimates thereof (solid lines) and their corresponding minima/roots (dots on the lines) in the setting described in Figure 1 using $\ell_1$-regularization: (a) $\text{DP}(\alpha, Ax^*)$ and $\text{DP}(\alpha, y^k)$. (b) $\text{MSPE}(\alpha)$ (empirical mean over $N_\varepsilon = 10^4$) and $\text{PSURE}(\alpha, y^k)$. (c) $\text{MSEE}(\alpha)$ (empirical mean over $N_\varepsilon = 10^4$) and $\text{SURE}(\alpha, y^k)$
Empirical probabilities of (a) $\alpha$ and (b) the corresponding $\ell_1$-error for different parameter choice rules using $\ell_1$-regularization, $m = n = 64$, $l = 0.06$, $\sigma = 0.1$ and $N = 10^4$ samples of $\varepsilon$
Illustration that Theorems 3.3 and 3.8 might also hold for $\ell_1$-regularization: The left hand side of (22)/(23) is estimated by the sample mean and plotted vs. $m$. The black dotted lines were added to compare the order of convergence
Illustration of the difficulties of evaluating the SURE risk in the case of $\ell_1$-regularization: In (a), a coarse linear grid is constructed around ${{\hat{\alpha }}_{\text{DP}}}$ as $\alpha = \Delta_\alpha, 2 \Delta_\alpha, \ldots, 20\Delta_\alpha$ with $\Delta_\alpha = {{\hat{\alpha }}_{\text{DP}}}/10$. Similar to Figure 11(a) the plot suggests a clear minimum. However, using a fine, logarithmic grid, (b) reveals that it is only a sub-optimal local minimum before a very erratic part of $\text{SURE}(\alpha, y)$ starts. (c) shows how a coarse $\alpha$-grid can lead to an arbitrary projection of $\text{SURE}(\alpha, y)$ that is likely to miss important features. Both risk estimators are the same as those plotted in Figure 12(c)with the same colors. In (d), the difference between computing $\text{SURE}(\alpha, y)$ with the consistent and highly accurate version of ADMM (Impl A) and with a standard ADMM version using only 20 iterations (Impl B) is illustrated
Condition of $A_l$ computed different values of $m = n$ and $l$
 $l = 0.02$ $l=0.04$ $l=0.06$ $l=0.08$ $l=0.1$ $m=16$ 1.27e+0 1.75e+0 2.79e+0 6.77e+0 2.31e+2 $m=32$ 1.75e+0 6.77e+0 6.94e+1 6.88e+2 2.30e+2 $m=64$ 6.77e+0 6.88e+2 6.42e+2 1.51e+3 4.22e+3 $m=128$ 6.88e+2 1.51e+3 1.51e+4 4.29e+3 4.29e+4 $m=256$ 1.70e+3 4.70e+4 1.87e+6 4.07e+6 1.79e+6 $m=512$ 4.70e+4 1.11e+7 1.22e+7 2.12e+7 3.70e+7
 $l = 0.02$ $l=0.04$ $l=0.06$ $l=0.08$ $l=0.1$ $m=16$ 1.27e+0 1.75e+0 2.79e+0 6.77e+0 2.31e+2 $m=32$ 1.75e+0 6.77e+0 6.94e+1 6.88e+2 2.30e+2 $m=64$ 6.77e+0 6.88e+2 6.42e+2 1.51e+3 4.22e+3 $m=128$ 6.88e+2 1.51e+3 1.51e+4 4.29e+3 4.29e+4 $m=256$ 1.70e+3 4.70e+4 1.87e+6 4.07e+6 1.79e+6 $m=512$ 4.70e+4 1.11e+7 1.22e+7 2.12e+7 3.70e+7
Statistics of the $\ell_2$-error $\|x^* - x_{\hat \alpha}\|_2$ for different parameter choice rules using $m = n = 64$, $l = 0.06$, $\sigma = 0.1$ and $N_\varepsilon = 10^6$ samples of $\varepsilon$
 min max mean median std optimal 4.78 9.63 8.04 8.05 0.43 DP 6.57 10.81 8.82 8.87 0.34 PSURE 6.10 277.24 8.38 8.23 1.53 SURE 6.08 339.80 27.71 8.95 37.26
 min max mean median std optimal 4.78 9.63 8.04 8.05 0.43 DP 6.57 10.81 8.82 8.87 0.34 PSURE 6.10 277.24 8.38 8.23 1.53 SURE 6.08 339.80 27.71 8.95 37.26
 [1] Stefan Kindermann. Convergence of the gradient method for ill-posed problems. Inverse Problems & Imaging, 2017, 11 (4) : 703-720. doi: 10.3934/ipi.2017033 [2] Matthew A. Fury. Regularization for ill-posed inhomogeneous evolution problems in a Hilbert space. Conference Publications, 2013, 2013 (special) : 259-272. doi: 10.3934/proc.2013.2013.259 [3] Paola Favati, Grazia Lotti, Ornella Menchi, Francesco Romani. An inner-outer regularizing method for ill-posed problems. Inverse Problems & Imaging, 2014, 8 (2) : 409-420. doi: 10.3934/ipi.2014.8.409 [4] Vinicius Albani, Adriano De Cezaro, Jorge P. Zubelli. On the choice of the Tikhonov regularization parameter and the discretization level: A discrepancy-based strategy. Inverse Problems & Imaging, 2016, 10 (1) : 1-25. doi: 10.3934/ipi.2016.10.1 [5] Matthew A. Fury. Estimates for solutions of nonautonomous semilinear ill-posed problems. Conference Publications, 2015, 2015 (special) : 479-488. doi: 10.3934/proc.2015.0479 [6] Guozhi Dong, Bert Jüttler, Otmar Scherzer, Thomas Takacs. Convergence of Tikhonov regularization for solving ill-posed operator equations with solutions defined on surfaces. Inverse Problems & Imaging, 2017, 11 (2) : 221-246. doi: 10.3934/ipi.2017011 [7] Olha P. Kupenko, Rosanna Manzo. On optimal controls in coefficients for ill-posed non-Linear elliptic Dirichlet boundary value problems. Discrete & Continuous Dynamical Systems - B, 2018, 23 (4) : 1363-1393. doi: 10.3934/dcdsb.2018155 [8] Sergiy Zhuk. Inverse problems for linear ill-posed differential-algebraic equations with uncertain parameters. Conference Publications, 2011, 2011 (Special) : 1467-1476. doi: 10.3934/proc.2011.2011.1467 [9] Eliane Bécache, Laurent Bourgeois, Lucas Franceschini, Jérémi Dardé. Application of mixed formulations of quasi-reversibility to solve ill-posed problems for heat and wave equations: The 1D case. Inverse Problems & Imaging, 2015, 9 (4) : 971-1002. doi: 10.3934/ipi.2015.9.971 [10] Markus Haltmeier, Richard Kowar, Antonio Leitão, Otmar Scherzer. Kaczmarz methods for regularizing nonlinear ill-posed equations II: Applications. Inverse Problems & Imaging, 2007, 1 (3) : 507-523. doi: 10.3934/ipi.2007.1.507 [11] Misha Perepelitsa. An ill-posed problem for the Navier-Stokes equations for compressible flows. Discrete & Continuous Dynamical Systems - A, 2010, 26 (2) : 609-623. doi: 10.3934/dcds.2010.26.609 [12] Alina Toma, Bruno Sixou, Françoise Peyrin. Iterative choice of the optimal regularization parameter in TV image restoration. Inverse Problems & Imaging, 2015, 9 (4) : 1171-1191. doi: 10.3934/ipi.2015.9.1171 [13] Youri V. Egorov, Evariste Sanchez-Palencia. Remarks on certain singular perturbations with ill-posed limit in shell theory and elasticity. Discrete & Continuous Dynamical Systems - A, 2011, 31 (4) : 1293-1305. doi: 10.3934/dcds.2011.31.1293 [14] Johann Baumeister, Barbara Kaltenbacher, Antonio Leitão. On Levenberg-Marquardt-Kaczmarz iterative methods for solving systems of nonlinear ill-posed equations. Inverse Problems & Imaging, 2010, 4 (3) : 335-350. doi: 10.3934/ipi.2010.4.335 [15] Alfredo Lorenzi, Luca Lorenzi. A strongly ill-posed integrodifferential singular parabolic problem in the unit cube of $\mathbb{R}^n$. Evolution Equations & Control Theory, 2014, 3 (3) : 499-524. doi: 10.3934/eect.2014.3.499 [16] Faker Ben Belgacem. Uniqueness for an ill-posed reaction-dispersion model. Application to organic pollution in stream-waters. Inverse Problems & Imaging, 2012, 6 (2) : 163-181. doi: 10.3934/ipi.2012.6.163 [17] Markus Haltmeier, Antonio Leitão, Otmar Scherzer. Kaczmarz methods for regularizing nonlinear ill-posed equations I: convergence analysis. Inverse Problems & Imaging, 2007, 1 (2) : 289-298. doi: 10.3934/ipi.2007.1.289 [18] Adriano De Cezaro, Johann Baumeister, Antonio Leitão. Modified iterated Tikhonov methods for solving systems of nonlinear ill-posed equations. Inverse Problems & Imaging, 2011, 5 (1) : 1-17. doi: 10.3934/ipi.2011.5.1 [19] Lianwang Deng. Local integral manifolds for nonautonomous and ill-posed equations with sectorially dichotomous operator. Communications on Pure & Applied Analysis, 2020, 19 (1) : 145-174. doi: 10.3934/cpaa.2020009 [20] Peter I. Kogut, Olha P. Kupenko. On optimal control problem for an ill-posed strongly nonlinear elliptic equation with $p$-Laplace operator and $L^1$-type of nonlinearity. Discrete & Continuous Dynamical Systems - B, 2019, 24 (3) : 1273-1295. doi: 10.3934/dcdsb.2019016

2018 Impact Factor: 1.469