betaFeedback

WorksMathematics

Subgradient Methods for Nonsmooth Sample Average Approximation

Abstract

We study direct subgradient methods for nonsmooth sample average approximation (SAA). Subgradient descent and Nesterov acceleration with constant step sizes attain polynomial and geometric rates matching their smooth counterparts. After O(log⁡n)O(\log n) full-batch updates, where nn is the sample size, the computed solution has the same first-order expansion and limiting distribution as an SAA solution, even when the iterates do not converge. A bound on subgradient increments transfers population regularity to the sample objective, controls the algorithmic trajectory, and identifies the leading random term. The bound covers convex piecewise linear scenario costs, including two-stage stochastic linear programs with fixed recourse. Moreau smoothing displaces the population solution at first order and changes the limiting distribution of the computed solution. An explicit shift of the residual removes the first-order displacement and improves the expansion remainder. For sparse decisions, restricted increment bounds give convergence, support recovery, and expansions for proximal subgradient descent and iterative hard thresholding.

Introduction

Sample average approximation (SAA) replaces an expected cost by an average over observed scenarios. The accuracy of a computed solution depends on both sampling error and optimization error. We study how quickly direct subgradient methods reduce the optimization error to the scale needed for inference when the scenario costs are nonsmooth.

Averaging piecewise linear costs against a distribution with a bounded conditional density produces a smooth expected objective. We show how its local strong convexity speeds computation on the nonsmooth sample objective and determines the sampling distribution of the computed solution.

Our main probabilistic result relates subgradient increments to the distance between decisions. Consider L(θ)=n−1∑if(Yi−ZiTθ)+bTθ\mathcal{L}(\theta)=n^{-1}\sum_i f(Y_i-Z_i^{T}\theta)+b^{T}\theta on Θ⊆Rp\Theta\subseteq\mathbb{R}^{p}, where ff is convex and Lipschitz. Let L‾\overline{\mathcal{L}} be its expected objective, write ∂L\partial\mathcal{L} for a selected subgradient, and set an=plog⁡n/na_n=p\log n/n. We prove

∥∂L(θ1)−∂L(θ2)−∇L‾(θ1)+∇L‾(θ2)∥2≤C{an∥θ1−θ2∥2+an}\left\lVert\partial\mathcal{L}(\theta_1)-\partial\mathcal{L}(\theta_2)-\nabla\overline{\mathcal{L}}(\theta_1)+\nabla\overline{\mathcal{L}}(\theta_2)\right\rVert_2 \le C\left\{\sqrt{a_n}\lVert\theta_1-\theta_2\rVert_2+a_n\right\}

with high probability, simultaneously for every pair (θ1,θ2)∈Θ×Θ(\theta_1,\theta_2)\in\Theta\times\Theta. A pinball example shows that the smoothness approximation scale ana_n is necessary up to a logarithmic factor. The assumptions control the conditional density of YY and projections of ZZ, without a moment condition on YY. The same bound extends to finite maxima of random affine costs, including two-stage stochastic linear programs with fixed recourse.

The increment bound controls successive iterates, extrapolated points, and the SAA solution on a common event, although all depend on the same observations. At distances larger than ana_n, transfers smoothness and strong convexity from the expected objective to approximate properties of the SAA objective. Near the population solution, the distance dependence gives a linear subgradient approximation that identifies the random term governing inference.

For Lipschitz convex objectives, subgradient descent with step sizes proportional to t−ζt^{-\zeta}, 1/2<ζ<11/2 < \zeta< 1, attains objective error O(T−1)O(T^{-1}) for a weighted average. Under strong convexity, diminishing steps and averaging give O(T−1)O(T^{-1}) ([32]). Our increment bound gives three phases of faster convergence for SAA. In Phase 1, subgradient descent and Nesterov acceleration attain objective rates T−1T^{-1} and T−2T^{-2} and enter a region of approximate strong convexity. In Phase 2, their errors decrease toward approximation floors at rates exp⁡(−cT/κ)\exp(-cT/\kappa) and exp⁡(−cT/κ)\exp(-cT/\sqrt{\kappa}), where κ\kappa is the ratio of the smoothness and strong convexity parameters. A one-dimensional example shows that, from almost every initialization, subgradient descent cannot reduce its error below order ϵ1\epsilon_1, the smoothness approximation scale. Phase 3 separates the remaining optimization error from the leading random term. Acceleration improves the geometric rate while preserving the expansion remainder.

These convergence bounds complement analyses based on linear growth and sharpness ([44, 12, 17]). The two-phase method of Shen et al. [53] uses decreasing step sizes initially and a constant step size in its second phase. Our methods use a constant step size throughout and yield a distributional approximation of the computed solution. Beyond SAA, bounded local subgradient variation and quadratic growth give geometric convergence of subgradient descent and restarted Nesterov acceleration to neighborhoods whose radii have the order of the variation scale.

The iterates need not converge to a minimizer to support inference. After O(log⁡n)O(\log n) updates, their optimization error is O(an)O(a_n), and the computed solution has the same first-order expansion as an SAA solution, with remainder OP{(p/n)3/4(log⁡n)1/2+an}O_P\{(p/n)^{3/4}(\log n)^{1/2}+a_n\}. First-order equivalence holds under weaker dimension growth conditions than those needed for a Gaussian limit. The same analysis gives bounds on the population optimality gap and the sample complexity of an ϵ\epsilon-optimal decision. Each update evaluates nn scenario subgradients, giving a total of O(nlog⁡n)O(n\log n) scenario evaluations.

Moreau smoothing can change the computed solution’s limiting distribution through a drift nh vTBw\sqrt{n}h\,v^\mathsf{T}B_w in a unit contrast vv. We identify the first-order population displacement coefficient BwB_w for losses with two affine pieces and general prox-functions. We derive a residual shift hcwh_{c_w} that removes the first-order displacement. The expansion remainder then improves to OP{(p/n)4/5(log⁡n)2/5}O_P\{(p/n)^{4/5}(\log n)^{2/5}\}, with the limiting distribution preserved. Sample averaging keeps the smoothness and local strong convexity constants bounded as the bandwidth decreases to CanC a_n, permitting constant step sizes and convergence to the smoothed SAA solution.

Sparse decisions lead to regularized SAA. Restricting the increments to sparse vectors gives an approximation scale slog⁡p/ns\log p/n when the working sparsity is proportional to the population sparsity ss. For proximal subgradient descent and iterative hard thresholding, this yields geometric convergence, support recovery, and expansions of the computed solution. On a recovered support or a deterministic retained set containing the population support, the expansions give coordinatewise normal limits after O(log⁡n)O(\log n) updates.

We determine the iteration budget needed to retain the sampling distribution described by the asymptotic theory of SAA solutions ([29, 51, 52]). When fewer updates are used, the expansion also quantifies the remaining optimization error. Retrospective approximation and sample-allocation analyses use an inner solver’s convergence rate to choose sampling and stopping schedules ([43, 48]). We derive that rate for the nonsmooth SAA solver and relate its stopping accuracy to the solution expansion. The same result gives a full-batch counterpart to asymptotic normality for stochastic approximation with iterate averaging ([50, 45 Davis et al.,]) 2024). Here the increment bound controls dependence on the repeatedly used sample and gives an expansion of the final iterate.

The increment bound supplies both the optimization models and the probability control needed for this expansion. The objective recursions follow the inexact-oracle framework [14, 15], whose accelerated bounds accumulate model error with iterations. Here long updates remove that error, and we prove containment of iterates and extrapolated points within the regions where the models hold. The resulting approximation scales also give high-probability values of the regularity parameters in universal-gradient and local-variation analyses [39, 16].

To pass from optimization bounds to an expansion, we use the decrease of the increment error with distance. Localization of objective deviations over near-optimal sets gives bounds on SAA solutions [41 b]. Other nonsmooth SAA analyses give concentration of subdifferentials and Moreau envelopes [11]. In particular, Ruan [49] establishes uniform Hausdorff convergence of empirical subdifferentials. Subtracting two such approximation bounds retains their full error at nearby decisions. Our direct control of increments makes the error decrease as the decisions approach each other, which gives the smaller remainder in the computed solution’s expansion.

Organization and notation

Section 2 states the regularity conditions. Section 3 gives the convergence bounds and their combination. Section 4 verifies the conditions for SAA objectives and develops applications. Sections 5 and 6 treat smoothing and regularized SAA. Proofs are collected in the Appendix.

For a vector v∈Rpv \in\mathbb{R}^{p}, we use ∣v∣2|v|_{2}, ∣v∣1|v|_{1}, and ∣v∣∞|v|_{\infty} for its Euclidean, ℓ1\ell_{1}, and maximum norms. For an index set II, vIv_{I} is the subvector indexed by II, and v−lv_{-l} omits coordinate ll. For a matrix AA, ∥A∥\lVert A \rVert and ∥A∥∞\lVert A \rVert_{\infty} are the induced Euclidean and maximum norms. For a symmetric matrix, Λmax⁡(A)\Lambda_{\max}(A) and Λmin⁡(A)\Lambda_{\min}(A) are its largest and smallest eigenvalues. We write AI×JA_{I \times J} for the submatrix indexed by rows II and columns JJ. The indicator I(A)\mathbb{I}(A) equals one when the event or condition AA holds and zero otherwise. We write B(θ,r)B(\theta,r) for the closed Euclidean ball with center θ\theta and radius rr. For a random variable WW and q>0q>0, write ∥W∥ψq=inf⁡{c>0:Eexp⁡(∣W/c∣q)≤2}\lVert W \rVert_{\psi_{q}} = \inf\{c>0:\mathbb{E}\exp(|W/c|^{q}) \le2\}. For positive sequences, an≍bna_{n} \asymp b_{n} means that their ratio is bounded above and away from zero. Unless stated otherwise, constants and the step sizes used in asymptotic iteration bounds are independent of nn and pp.

Problem and regularity conditions

Let L\mathcal{L} be an SAA objective on the decision region Θ⊆Rp\Theta\subseteq\mathbb{R}^{p}, and let L~\widetilde{\mathcal{L}} be its expected objective. When individual losses are not integrable, use the loss-difference definition in Section 4. We write

θ∗∈argmin⁡θ∈Θ L~(θ),θ^∈argmin⁡θ∈Θ L(θ)(1)\theta^{*} \in\underset{\theta\in\Theta}{\operatorname{argmin}}\,\widetilde{\mathcal{L}}(\theta), \qquad\widehat{\theta} \in\underset{\theta\in\Theta}{\operatorname{argmin}}\,\mathcal{L}(\theta) \tag*{(1)}

for the population solution and an SAA solution. For a computed solution θ~\widetilde{\theta}, the differences θ~−θ^\widetilde{\theta}-\widehat{\theta} and θ^−θ∗\widehat{\theta}-\theta^{*} are the optimization error and sampling error, respectively. The sample remains fixed throughout the algorithm. We use ∂L(θ)\partial\mathcal{L}(\theta) for a measurable selection of a subgradient at θ\theta. At a differentiable point it equals ∇L(θ)\nabla\mathcal{L}(\theta). All updates and derivatives are taken in the free decision coordinates. Affine feasible sets are parametrized by these coordinates.

We first state deterministic conditions under which subgradient descent and Nesterov acceleration inherit convergence properties of smooth optimization. Section 4 verifies these conditions with high probability for a class of SAA objectives.

Approximate smoothness and strong convexity

Condition 1. For L>0L > 0 and ϵ1≥0\epsilon_1 \ge0, the function L\mathcal{L} is (L,ϵ1)(L,\epsilon_1)-smooth on a set UU if, for all θ1,θ2∈U\theta_1,\theta_2 \in U,

∥∂L(θ1)−∂L(θ2)∥2≤Lmax⁡{∥θ1−θ2∥2,ϵ1}.\lVert\partial\mathcal{L}(\theta_1)-\partial\mathcal{L}(\theta_2)\rVert_2 \le L\max\{\lVert\theta_1-\theta_2\rVert_2,\epsilon_1\}.

Condition 1 bounds subgradient increments by distance when that distance is at least ϵ1\epsilon_1, and by Lϵ1L\epsilon_1 at smaller distances. This permits discontinuities in the selected subgradient while retaining the quadratic upper bound used in descent analysis. Every function with maximum local subgradient variation bounded by Vϵ1V_{\epsilon_1} at radius ϵ1>0\epsilon_1 > 0 on a convex region satisfies Condition 1 with L=2Vϵ1/ϵ1L = 2V_{\epsilon_1}/\epsilon_1. Partitioning a segment into pieces of length at most ϵ1\epsilon_1 gives this implication from the BVGmax⁡_{\max} class of Diakonikolas and Guzmán [16].

Lemma 1. Suppose Condition 1 holds for a finite convex function L\mathcal{L} on a convex region. For every pair θ1,θ2\theta_1,\theta_2 in that region with ∥θ1−θ2∥2≥ϵ1\lVert\theta_1-\theta_2\rVert_2 \ge\epsilon_1,

L(θ1)−L(θ2)≤⟨∂L(θ2),θ1−θ2⟩+L∥θ1−θ2∥22.(2)\mathcal{L}(\theta_1)-\mathcal{L}(\theta_2) \le\langle\partial\mathcal{L}(\theta_2),\theta_1-\theta_2\rangle+L\lVert\theta_1-\theta_2\rVert_2^2. \tag*{(2)}

For 0<η≤1/L0 < \eta\le1/L and θ,θ+=θ−η∂L(θ)\theta,\theta^+ = \theta-\eta\partial\mathcal{L}(\theta) in the region,

L(θ+)≤L(θ)−η2∥∂L(θ)∥22+ηL2ϵ122.(3)\mathcal{L}(\theta^+) \le\mathcal{L}(\theta)-\frac{\eta}{2}\lVert\partial\mathcal{L}(\theta)\rVert_2^2+\frac{\eta L^2\epsilon_1^2}{2}. \tag*{(3)}

The descent bound (3) applies at every update. For η≤1/(2L)\eta\le1/(2L), an update of length at least ϵ1\epsilon_1 removes the additive error by (2).

Condition 2. For μ>0\mu> 0 and ϵ2≥0\epsilon_2 \ge0, the function L\mathcal{L} is (μ,ϵ2)(\mu,\epsilon_2)-strongly convex on a set UU if

L(θ1)−L(θ2)−⟨∂L(θ2),θ1−θ2⟩≥μ4∥θ1−θ2∥22\mathcal{L}(\theta_1)-\mathcal{L}(\theta_2)-\langle\partial\mathcal{L}(\theta_2),\theta_1-\theta_2\rangle\ge\frac{\mu}{4}\lVert\theta_1-\theta_2\rVert_2^2

for all θ1,θ2∈U\theta_1,\theta_2 \in U with ∥θ1−θ2∥2≥ϵ2\lVert\theta_1-\theta_2\rVert_2 \ge\epsilon_2.

References

  1. [1]Alekh Agarwal, Sahand Negahban, and Martin J. Wainwright. Fast global convergence of gradient methods for high-dimensional statistical recovery. The Annals of Statistics, 40(5):2452–2482, 2012.
  2. [2]Gah-Yi Ban and Cynthia Rudin. The big data newsvendor: Practical insights from machine learning. Operations Research, 67(1):90–108, 2019.
  3. [3]Amir Beck. First-order methods in optimization. Society for Industrial and Applied Mathematics, Philadelphia, PA, 2017.DOI
  4. [4]Amir Beck and Marc Teboulle. Smoothing and first order methods: a unified framework. SIAM Journal on Optimization, 22(2):557–580, 2012.DOI
  5. [5]Thomas Blumensath and Mike E. Davies. Iterative hard thresholding for compressed sensing. Applied and Computational Harmonic Analysis, 27(3):265–274, 2009.arxiv.org/abs/0805.0510
  6. [6]Florentina Bunea, Alexandre B. Tsybakov, and Marten H. Wegkamp. Aggregation for Gaussian regression. The Annals of Statistics, 35(4):1674–1697, 2007.
  7. [7]Jinyuan Chang, Lin Yang, Yichen Zhang, and Wen-Xin Zhou. Feature-rich, data-private: a sparse learning framework for the high-dimensional newsvendor. Operations Research, 2026. Articles in Advance.
  8. [8]Le-Yu Chen and Sokbae Lee. Sparse quantile regression. Journal of Econometrics, 235(2):2195–2217, 2023.
  9. [9]Xi Chen, Weidong Liu, and Yichen Zhang. Quantile regression under memory constraint. The Annals of Statistics, 47(6):3244–3273, 2019.DOI
  10. [10]Victor Chernozhukov, Denis Chetverikov, Kengo Kato, and Yuta Koike. Improved central limit theorem and bootstrap approximations in high dimensions. The Annals of Statistics, 50(5):2562–2586, 2022.arxiv.org/abs/1912.10529
  11. [11]Damek Davis and Dmitriy Drusvyatskiy. Graphical convergence of subgradients in nonconvex optimization and learning. Mathematics of Operations Research, 47(1):209–231, 2022.arxiv.org/abs/1810.07590
  12. [12]Damek Davis, Dmitriy Drusvyatskiy, Kellie J. MacPhee, and Courtney Paquette. Subgradient methods for sharp weakly convex functions. Journal of Optimization Theory and Applications, 179:962–982, 2018.arxiv.org/abs/1803.02461
  13. [14]Olivier Devolder, François Glineur, and Yurii Nesterov. First-order methods with inexact oracle: the strongly convex case. CORE Discussion Paper 2013/16, CORE, 2013.
  14. [15]Olivier Devolder, François Glineur, and Yurii Nesterov. First-order methods of smooth convex optimization with inexact oracle. Mathematical Programming, 146(1–2):37–75, 2014.
  15. [16]Jelena Diakonikolas and Cristóbal Guzmán. Optimization on a finer scale: bounded local subgradient variation perspective. SIAM Journal on Optimization, 36(1):152–184, 2026.arxiv.org/abs/2403.16317
  16. [17]Lijun Ding and Alex L. Wang. Sharpness and conditioning of nonsmooth convex formulations in statistical signal recovery. SIAM Journal on Optimization, 36(3):1269–1296, 2026.
  17. [18]Asen L. Dontchev and R. Tyrrell Rockafellar. Regularity and conditioning of solution mappings in variational analysis. Set-Valued Analysis, 12:79–109, 2004.DOI
  18. [19]John C. Duchi, Peter L. Bartlett, and Martin J. Wainwright. Randomized smoothing for stochastic optimization. SIAM Journal on Optimization, 22(2):674–701, 2012.
  19. [20]Marcelo Fernandes, Emmanuel Guerre, and Eduardo Horta. Smoothing quantile regressions. Journal of Business & Economic Statistics, 39(1):338–357, 2021.
  20. [21]Xuming He and Qi-Man Shao. A general Bahadur representation of M-estimators and its application to linear regression with nonstochastic designs. The Annals of Statistics, 24(6):2608–2630, 1996.DOI
  21. [22]Xuming He, Xiaolou Pan, Kean Ming Tan, and Wen-Xin Zhou. Smoothed quantile regression with large-scale inference. Journal of Econometrics, 232(2):367–388, 2023a.
  22. [23]Xuming He, Kean Ming Tan, and Wen-Xin Zhou. Robust estimation and inference for expected shortfall regression with many regressors. Journal of the Royal Statistical Society: Series B, 85(4):1223–1246, 2023b.arxiv.org/abs/2212.05565
  23. [24]Joel L. Horowitz. Bootstrap methods for median regression models. Econometrica, 66(6):1327–1351, 1998.DOI
  24. [25]Louis A. Jaeckel. Estimating regression coefficients by minimizing the dispersion of the residuals. The Annals of Mathematical Statistics, 43(5):1449–1458, 1972.DOI
  25. [26]Prateek Jain, Ambuj Tewari, and Purushottam Kar. On iterative hard thresholding methods for high-dimensional M-estimation. In Advances in Neural Information Processing Systems, 2014.arxiv.org/abs/1410.5137
  26. [27]Tobias Lindström Jensen, Jakob Heide Jørgensen, Per Christian Hansen, and Søren Holdt Jensen. Implementation of an optimal first-order method for strongly convex total variation regularization. BIT Numerical Mathematics, 52:329–356, 2012.
  27. [28]Billy Jin, Katya Scheinberg, and Miaolan Xie. High probability complexity bounds for adaptive step search based on stochastic oracles. SIAM Journal on Optimization, 34(3):2411–2439, 2024.arxiv.org/abs/2106.06454
  28. [29]Alan J. King and R. Tyrrell Rockafellar. Asymptotic theory for solutions in statistical estimation and stochastic programming. Mathematics of Operations Research, 18(1):148–162, 1993.DOI
  29. [30]Roger Koenker. Quantile regression. Cambridge University Press, 2005.
  30. [31]Ja-Yong Koo, Yoonkyung Lee, Yuwon Kim, and Changyi Park. A Bahadur representation of the linear support vector machine. Journal of Machine Learning Research, 9(44):1343–1368, 2008.
  31. [32]Simon Lacoste-Julien, Mark Schmidt, and Francis Bach. A simpler approach to obtaining an O(1/t) convergence rate for the projected stochastic subgradient method, 2012.
  32. [33]Henry Lam, Kaizheng Wang, Yuhang Wu, and Yichen Zhang. Adaptive data fusion for multi-task non-smooth optimization. arXiv preprint arXiv:2210.12334, 2022.arxiv.org/abs/2210.12334
  33. [34]J. J. Moreau. Proximité et dualité dans un espace hilbertien. Bulletin de la Société Mathématique de France, 93:273–299, 1965.DOI
  34. [35]Fedor Nazarov. On the maximal perimeter of a convex set in ℝⁿ with respect to a Gaussian measure. In Vitali D. Milman and Gideon Schechtman, editors, Geometric Aspects of Functional Analysis, volume 1807 of Lecture Notes in Mathematics, pages 169–187. Springer, 2003.DOI
  35. [36]Sahand Negahban, Bin Yu, Martin J. Wainwright, and Pradeep Ravikumar. A unified framework for high-dimensional analysis of M-estimators with decomposable regularizers. In Advances in Neural Information Processing Systems, 2009.
  36. [37]Yurii Nesterov. A method for solving the convex programming problem with convergence rate o(1/k²). Dokl. Akad. Nauk SSSR, 269(3):543–547, 1983.
  37. [38]Yurii Nesterov. Smooth minimization of non-smooth functions. Mathematical Programming, 103(1):127–152, 2005.
  38. [39]Yurii Nesterov. Universal gradient methods for convex optimization problems. Mathematical Programming, 152(1–2):381–404, 2015.DOI
  39. [40]Brendan O’Donoghue and Emmanuel J. Candès. Adaptive restart for accelerated gradient schemes. Foundations of Computational Mathematics, 15:715–732, 2015.
  40. [41]Roberto I. Oliveira and Philip Thompson. Sample average approximation with heavier tails I: non-asymptotic bounds with weak assumptions and stochastic constraints. Mathematical Programming, 199(1–2):1–48, 2023a.arxiv.org/abs/1705.00822
  41. [43]Raghu Pasupathy. On choosing parameters in retrospective-approximation algorithms for stochastic root finding and simulation optimization. Operations Research, 58(4):889–901, 2010.DOI
  42. [44]Boris T. Polyak. Minimization of unsmooth functionals. USSR Computational Mathematics and Mathematical Physics, 9(3):14–29, 1969.DOI
  43. [45]Boris T. Polyak and Anatoli B. Juditsky. Acceleration of stochastic approximation by averaging. SIAM Journal on Control and Optimization, 30(4):838–855, 1992.DOI
  44. [46]R. Tyrrell Rockafellar. Convex analysis, volume 28. Princeton University Press, 1997.
  45. [47]R. Tyrrell Rockafellar and Stanislav Uryasev. Optimization of conditional value-at-risk. The Journal of Risk, 2(3):21–41, 2000.DOI
  46. [48]Johannes O. Royset and Roberto Szechtman. Optimal budget allocation for sample average approximation. Operations Research, 61(3):762–776, 2013.
  47. [49]Feng Ruan. On the uniform convergence of subdifferentials in stochastic optimization and learning. Mathematics of Operations Research, 51(3):2323–2353, 2025.arxiv.org/abs/2405.10289
  48. [50]D. Ruppert. Efficient estimations from a slowly convergent Robbins–Monro process. Technical Report 781, School of Operations Research and Industrial Engineering, Cornell University, February 1988.
  49. [51]Alexander Shapiro. Asymptotic behavior of optimal solutions in stochastic programming. Mathematics of Operations Research, 18(4):829–845, 1993.DOI
  50. [52]Alexander Shapiro, Darinka Dentcheva, and Andrzej Ruszczyński. Lectures on Stochastic Programming: Modeling and Theory. Society for Industrial and Applied Mathematics, 3rd edition, 2021.
  51. [53]Yinan Shen, Jingyang Li, Jian-Feng Cai, and Dong Xia. Computationally efficient and statistically optimal robust high-dimensional linear regression. The Annals of Statistics, 53(1):374–399, 2025.arxiv.org/abs/2305.06199
  52. [54]Kean Ming Tan, Lan Wang, and Wen-Xin Zhou. High-dimensional quantile regression: convolution smoothing and concave regularization. Journal of the Royal Statistical Society: Series B, 84(1):205–233, 2022.arxiv.org/abs/2109.05640
  53. [55]Roman Vershynin. High-Dimensional Probability: An Introduction with Applications in Data Science, vol- ume 47 of Cambridge Series in Statistical and Probabilistic Mathematics. Cambridge University Press, 2018.
  54. [56]Lan Wang and Xuming He. Analysis of global and local optima of regularized quantile regression in high dimensions: a subgradient approach. Econometric Theory, 40(2):233–277, 2024.
  55. [57]Lan Wang, Bo Peng, Jelena Bradic, Runze Li, and Yunan Wu. A tuning-free robust and efficient approach to high-dimensional regression. Journal of the American Statistical Association, 115(532):1700–1714, 2020.
  56. [58]Zhaoran Wang, Han Liu, and Tong Zhang. Optimal computational and statistical rates of convergence for sparse nonconvex learning problems. The Annals of Statistics, 42(6):2164–2201, 2014.DOI
  57. [59]Lin Xiao and Tong Zhang. A proximal-gradient homotopy method for the sparse least-squares problem. SIAM Journal on Optimization, 23(2):1062–1091, 2013.arxiv.org/abs/1203.3002

Paper details

vlnc:W_2cj7z5ev
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents