pith. sign in

arxiv: 2605.25034 · v1 · pith:E52QRAAGnew · submitted 2026-05-24 · 🧮 math.NA · cs.NA

Randomized conjugate gradient least squares

Yun Zeng , Jian-Feng Cai , Deren Han , Jiaxin Xie This is my paper

Pith reviewed 2026-06-29 23:47 UTC · model grok-4.3

classification 🧮 math.NA cs.NA
keywords randomized conjugate gradientleast squareslinear convergencevariance reductioncoordinate descentridge regressioniterative sketching
0
0 comments X

The pith

RCGLS converges linearly in expectation with a better bound than randomized coordinate descent for least squares

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper develops a randomized conjugate gradient least squares method that applies iterative sketching to cut the dimension and cost of each step when solving least-squares problems. It reinterprets classical CGLS by selecting the next descent direction through a constraint correction problem tied to the gradient. Replacing that gradient with a randomized coordinate gradient, which carries a variance reduction property, produces the new method directly. The authors prove linear convergence in expectation and establish a tighter bound than randomized coordinate descent achieves. They also supply an implementation that skips full-dimensional operations for sparse cases and show how the same idea solves ridge regression.

Core claim

By viewing the next descent direction in CGLS as the solution of a constraint correction problem associated with the gradient, the full gradient can be replaced by a randomized coordinate gradient that satisfies the variance reduction property. This replacement produces RCGLS, which is shown to converge linearly in expectation with a better convergence bound than the randomized coordinate descent method.

What carries the argument

The constraint correction formulation of the descent direction in CGLS, which allows direct substitution of the gradient by a randomized coordinate gradient while retaining the linear convergence property.

If this is right

  • RCGLS admits an implementation that avoids full-dimensional vector operations and is therefore practical for sparse matrices.
  • The same construction yields a lightweight parallelizable solver for ridge regression.
  • Numerical experiments confirm the linear expected convergence and the improved bound relative to coordinate descent.
  • Iterative sketching at each step reduces the working dimension while preserving the convergence guarantee.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The constraint-correction perspective on descent directions may allow similar randomizations inside other Krylov subspace methods.
  • Hybrid sampling strategies that mix coordinate gradients with other estimators could be tested for further acceleration on very large systems.
  • The variance-reduction property exploited here might extend to randomized estimators in related first-order optimization settings.

Load-bearing premise

The randomized coordinate gradient must satisfy the variance reduction property so it can replace the full gradient inside the constraint correction problem.

What would settle it

A counter-example least-squares problem where the expected convergence rate of RCGLS is no better than that of randomized coordinate descent would falsify the main claim.

Figures

Figures reproduced from arXiv: 2605.25034 by Deren Han, Jian-Feng Cai, Jiaxin Xie, Yun Zeng.

Figure 1
Figure 1. Figure 1: Figures depict the evolution of the number of epochs and the CPU time with respect to the block size q for synthetic datasets, with λ = 0.05. The title of each plot indicates the values of n and d. All computations are terminated once RSE < 10−10 . small, owing to the lower per-iteration computational cost of RidgeGRCDU compared with RidgeRCGLSU. Nevertheless, RidgeRCGLSU outperforms RidgeGRCDU in CPU time… view at source ↗
Figure 2
Figure 2. Figure 2: Figures depict the evolution of the number of epochs and the CPU time with respect to the block size q for synthetic datasets, with λ = 0.005. The title of each plot indicates the values of n and d. All computations are terminated once RSE < 10−10 . HImRidgeSketchU in terms of both flops and CPU time across all real-world datasets. In particular, on the protein dataset, RidgeRCGLSU converges much faster to… view at source ↗
Figure 3
Figure 3. Figure 3: Performance of HImRidgeSketchU and RidgeRCGLSU on syn￾thetic datasets with λ = 0.05. Figures depict the total flops and the CPU time across varying dimensions for overdetermined (n > d, left), square (n = d, middle), and underdetermined (n < d, right) cases. We set q = 50 and terminate the algorithms once RSE < 10−10. The titles specify the fixed parameters for each case. RCGLS can be equivalently reformul… view at source ↗
Figure 4
Figure 4. Figure 4: Performance of HImRidgeSketchU and RidgeRCGLSU on syn￾thetic datasets with λ = 0.005. Figures depict the total flops and the CPU time across varying dimensions for overdetermined (n > d, left), square (n = d, middle), and underdetermined (n < d, right) cases. We set q = 50 and terminate the algorithms once RSE < 10−10. The titles specify the fixed parameters for each case. RCGLS framework can be extended t… view at source ↗
Figure 5
Figure 5. Figure 5: Performance on overdetermined datasets from LIBSVM [5] with λ = 0.05. Figures depict the evolution of RSE with respect to the total flops and the CPU time. We set q = 2000 for real-sim and q = 50 for protein, and terminate the algorithms once RSE < 10−4 or the number of iterations exceeds a certain limit. The title of each plot indicates the names and sizes of the data. [7] Kui Du, Wu-Tao Si, and Xiao-Hui … view at source ↗
Figure 6
Figure 6. Figure 6: Performance on underdetermined datasets from LIBSVM [5] with λ = 0.05. Figures depict the evolution of RSE with respect to the total flops and the CPU time. We set q = 500 for LEDGAR and q = 50 for gisette, and terminate the algorithms once RSE < 10−4 . The title of each plot indicates the names and sizes of the data. [18] Per Christian Hansen. Regularization tools: A Matlab package for analysis and soluti… view at source ↗
read the original abstract

We develop a novel randomized conjugate gradient least squares (RCGLS) method for solving least-squares problems, in which iterative sketching is employed at each step to reduce the dimension and hence the computational cost. In particular, we propose a new perspective on the classical CGLS method, where the next descent direction is determined via a constraint correction problem associated with the gradient. Based on this insight, we replace the gradient with a randomized coordinate gradient that naturally satisfies the variance reduction property, leading directly to the proposed RCGLS method. We prove that RCGLS converges linearly in expectation, with a better convergence bound compared to the randomized coordinate descent method. Furthermore, we investigate an implementation of the method that avoids full-dimensional vector operations, which are the major bottleneck of vanilla RCGLS for sparse matrices and render it impractical. We also show how to apply the RCGLS method to solve the ridge regression problem, yielding a lightweight, parallelizable, and accelerated method for such problems. Numerical experiments are provided to confirm our results.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 0 minor

Summary. The paper proposes the randomized conjugate gradient least squares (RCGLS) method for least-squares problems. It derives the method from a new constraint-correction perspective on classical CGLS, replaces the full gradient with a randomized coordinate gradient that satisfies a variance-reduction property, proves linear convergence in expectation with a bound superior to that of randomized coordinate descent, develops an implementation avoiding full-dimensional operations for sparse matrices, extends the approach to ridge regression, and validates the results with numerical experiments.

Significance. If the claimed linear convergence result and improved bound hold under the stated assumptions, RCGLS would supply a theoretically stronger and practically implementable randomized solver for large-scale least squares and ridge regression, with potential advantages over coordinate-descent baselines in both rate and per-iteration cost for sparse data.

major comments (1)
  1. [Abstract] Abstract: the central claim of a linear convergence proof in expectation together with a strictly better bound than randomized coordinate descent is asserted without any visible derivation, theorem statement, assumption list, or error analysis; this absence prevents verification of the substitution step that replaces the gradient in the constraint-correction formulation.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their careful reading of the manuscript and for highlighting the need for clearer linkage between the abstract claims and the supporting analysis. We address the single major comment below.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim of a linear convergence proof in expectation together with a strictly better bound than randomized coordinate descent is asserted without any visible derivation, theorem statement, assumption list, or error analysis; this absence prevents verification of the substitution step that replaces the gradient in the constraint-correction formulation.

    Authors: The abstract is written as a concise summary of contributions and is not intended to contain full derivations or theorem statements, which is standard practice to respect length constraints. The constraint-correction view of CGLS, the substitution of the full gradient by a randomized coordinate gradient that satisfies the variance-reduction property, the complete theorem statement on linear convergence in expectation, the list of assumptions on the sketching matrices, the error analysis, and the explicit comparison showing a strictly better rate than randomized coordinate descent are all derived and proved in Section 3 (see in particular the statement and proof of Theorem 3.1 together with the surrounding discussion of the substitution step). We are happy to add a brief parenthetical reference in the abstract to the main theorem if the referee believes this would improve readability. revision: partial

Circularity Check

0 steps flagged

No significant circularity; derivation self-contained via stated substitution insight

full rationale

The paper derives RCGLS by viewing CGLS through a constraint-correction lens on the gradient, then substituting a randomized coordinate gradient (asserted to satisfy variance reduction by construction of the randomization). The linear convergence proof in expectation is claimed to follow directly from this replacement and yields a bound superior to RCD. No equations reduce a prediction to a fitted parameter by definition, no load-bearing self-citation chain is invoked to justify uniqueness or an ansatz, and the central claim does not rename a known result. The provided text and reader's assessment indicate an independent proof step rather than tautology.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Abstract provides no explicit free parameters or invented entities; the key premise is treated as a domain property of the randomized gradient.

axioms (1)
  • domain assumption The randomized coordinate gradient naturally satisfies the variance reduction property
    Invoked directly in the abstract to justify replacement of the gradient and derivation of RCGLS.

pith-pipeline@v0.9.1-grok · 5705 in / 1046 out tokens · 23977 ms · 2026-06-29T23:47:39.307935+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. On subspace-constrained preconditioning for randomized iterative methods

    math.NA 2026-05 unverdicted novelty 5.0

    Refines subspace preconditioning for randomized linear solvers via QR-like factorization, enabling implicit use and proving expected linear convergence while reducing to a smaller system with good singular values.

Reference graph

Works this paper leans on

55 extracted references · 4 canonical work pages · cited by 1 Pith paper · 1 internal anchor

  1. [1]

    Using ridge regression with genetic algorithm to enhance real estate appraisal forecasting

    Jae Joon Ahn, Hyun Woo Byun, Kyong Joo Oh, and Tae Yoon Kim. Using ridge regression with genetic algorithm to enhance real estate appraisal forecasting. Expert Syst. Appl. , 39(9):8369–8379, 2012

    2012

  2. [2]

    On partially randomized e xtended Kaczmarz method for solving large sparse overdetermined inconsistent linear systems

    Zhong-Zhi Bai and Wen-Ting Wu. On partially randomized e xtended Kaczmarz method for solving large sparse overdetermined inconsistent linear systems. Linear Algebra Appl. , 578:225–250, 2019

    2019

  3. [3]

    On greedy randomized augm ented Kaczmarz method for solving large sparse inconsistent linear systems

    Zhong-Zhi Bai and Wen-Ting Wu. On greedy randomized augm ented Kaczmarz method for solving large sparse inconsistent linear systems. SIAM J. Sci. Comput. , 43(6):A3892–A3911, 2021

    2021

  4. [4]

    SIAM, 2024

    ˚ Ake Bj¨ orck.Numerical methods for least squares problems . SIAM, 2024

    2024

  5. [5]

    LIBSVM: a library for support vector machines

    Chih-Chung Chang and Chih-Jen Lin. LIBSVM: a library for support vector machines. ACM Trans. Intell. Syst. Technol. , 2(3):1–27, 2011

    2011

  6. [6]

    Tight upper bounds for the convergence of the rand omized extended Kaczmarz and Gauss- Seidel algorithms

    Kui Du. Tight upper bounds for the convergence of the rand omized extended Kaczmarz and Gauss- Seidel algorithms. Numer. Linear Algebra Appl. , 26(3):e2233, 2019. RANDOMIZED CGLS 31 Figure 5. Performance on overdetermined datasets from LIBSVM [ 5] with λ = 0. 05. Figures depict the evolution of RSE with respect to the to tal flops and the CPU time. We set ...

    2019

  7. [7]

    Randomized extended average block Kaczmarz for solving least squares

    Kui Du, Wu-Tao Si, and Xiao-Hui Sun. Randomized extended average block Kaczmarz for solving least squares. SIAM J. Sci. Comput. , 42(6):A3541–A3559, 2020

    2020

  8. [8]

    Accelerated, para llel, and proximal coordinate descent

    Olivier Fercoq and Peter Richt´ arik. Accelerated, para llel, and proximal coordinate descent. SIAM J. Optim., 25(4):1997–2023, 2015

    1997

  9. [9]

    LSMR: An iter ative algorithm for sparse least-squares problems

    David Chin-Lung Fong and Michael Saunders. LSMR: An iter ative algorithm for sparse least-squares problems. SIAM J. Sci. Comput. , 33(5):2950–2971, 2011

    2011

  10. [10]

    Handbook of conv ergence theorems for (stochastic) gradient methods

    Guillaume Garrigos and Robert M Gower. Handbook of conv ergence theorems for (stochastic) gradient methods. arXiv preprint arXiv:2301.11235 , 2023

    work page arXiv 2023

  11. [11]

    Ri dgeSketch: a fast sketching based solver for large scale ridge regression

    Nidham Gazagnadou, Mark Ibrahim, and Robert M Gower. Ri dgeSketch: a fast sketching based solver for large scale ridge regression. SIAM J. Matrix Anal. Appl. , 43(3):1440–1468, 2022

    2022

  12. [12]

    Matrix computations

    Gene H Golub and Charles F Van Loan. Matrix computations. JHU press, 2013

    2013

  13. [13]

    Alge braic reconstruction techniques (ART) for three-dimensional electron microscopy and X-ray photogra phy

    Richard Gordon, Robert Bender, and Gabor T Herman. Alge braic reconstruction techniques (ART) for three-dimensional electron microscopy and X-ray photogra phy. J. Theor. Biol. , 29(3):471–481, 1970

    1970

  14. [14]

    Gower and Peter Richt´ arik

    Robert M. Gower and Peter Richt´ arik. Randomized itera tive methods for linear systems. SIAM J. Matrix Anal. Appl. , 36(4):1660–1690, 2015

    2015

  15. [15]

    Variance-reduced methods for machine learning

    Robert M Gower, Mark Schmidt, Francis Bach, and Peter Ri cht´ arik. Variance-reduced methods for machine learning. Proc. IEEE, 108(11):1968–1983, 2020

    1968

  16. [16]

    Randomized Doug las–Rachford methods for linear systems: Improved accuracy and efficiency

    Deren Han, Yansheng Su, and Jiaxin Xie. Randomized Doug las–Rachford methods for linear systems: Improved accuracy and efficiency. SIAM J. Optim. , 34(1):1045–1070, 2024

    2024

  17. [17]

    On pseudoinverse-free random ized methods for linear systems: Unified framework and acceleration

    Deren Han and Jiaxin Xie. On pseudoinverse-free random ized methods for linear systems: Unified framework and acceleration. Optim. Methods Softw. , 41(1):82–117, 2026. 32 YUN ZENG, JIAN-FENG CAI, DEREN HAN, AND JIAXIN XIE Figure 6. Performance on underdetermined datasets from LIBSVM [ 5] with λ = 0 . 05. Figures depict the evolution of RSE with respect to ...

    2026

  18. [18]

    Regularization tools: A Matlab p ackage for analysis and solution of discrete ill-posed problems

    Per Christian Hansen. Regularization tools: A Matlab p ackage for analysis and solution of discrete ill-posed problems. Numer. Algorithms , 6(1):1–35, 1994

    1994

  19. [19]

    Rows v ersus columns: Randomized Kaczmarz or Gauss-Seidel for ridge regression

    Ahmed Hefny, Deanna Needell, and Aaditya Ramdas. Rows v ersus columns: Randomized Kaczmarz or Gauss-Seidel for ridge regression. SIAM J. Sci. Comput. , 39(5):S528–S542, 2017

    2017

  20. [20]

    Algebraic reconstr uction techniques can be made computation- ally efficient (positron emission tomography application)

    Gabor T Herman and Lorraine B Meyer. Algebraic reconstr uction techniques can be made computation- ally efficient (positron emission tomography application). IEEE Trans. Med. Imaging , 12(3):600–609, 1993

    1993

  21. [21]

    Methods of con jugate gradients for solving linear systems

    Magnus R Hestenes, Eduard Stiefel, et al. Methods of con jugate gradients for solving linear systems. J. Res. Natl. Bur. Stand. , 49(6):409–436, 1952

    1952

  22. [22]

    Kaczmarz algorithm for Tikhonov regularization problem

    Andrey Aleksandrovich Ivanov and Aleksandr Ivanovich Zhdanov. Kaczmarz algorithm for Tikhonov regularization problem. Appl. Math. E-Notes , 13:270–276, 2013

    2013

  23. [23]

    Efficient accelerated coor dinate descent methods and faster algorithms for solving linear systems

    Yin Tat Lee and Aaron Sidford. Efficient accelerated coor dinate descent methods and faster algorithms for solving linear systems. In Proc. 54th Annu. IEEE Symp. Found. Comput. Sci. (FOCS) , pages 147–

  24. [24]

    Randomized method s for linear constraints: convergence rates and conditioning

    Dennis Leventhal and Adrian S Lewis. Randomized method s for linear constraints: convergence rates and conditioning. Math. Oper. Res. , 35(3):641–654, 2010

    2010

  25. [25]

    An accelerated ra ndomized proximal coordinate gradient method and its application to regularized empirical risk mi nimization

    Qihang Lin, Zhaosong Lu, and Lin Xiao. An accelerated ra ndomized proximal coordinate gradient method and its application to regularized empirical risk mi nimization. SIAM J. Optim. , 25(4):2244– 2273, 2015

    2015

  26. [26]

    An accelerated randomized Ka czmarz algorithm

    Ji Liu and Stephen Wright. An accelerated randomized Ka czmarz algorithm. Math. Comp., 85(297):153– 178, 2016. RANDOMIZED CGLS 33

    2016

  27. [27]

    Momentum and stoc hastic momentum for stochastic gradient, newton, proximal point and subspace descent methods

    Nicolas Loizou and Peter Richt´ arik. Momentum and stoc hastic momentum for stochastic gradient, newton, proximal point and subspace descent methods. Comput. Optim. Appl. , 77(3):653–710, 2020

    2020

  28. [28]

    Minimal error mom entum Bregman-Kaczmarz

    Dirk A Lorenz and Maximilian Winkler. Minimal error mom entum Bregman-Kaczmarz. Linear Algebra Appl., 2025

    2025

  29. [29]

    Convergen ce properties of the randomized extended Gauss–Seidel and Kaczmarz methods

    Anna Ma, Deanna Needell, and Aaditya Ramdas. Convergen ce properties of the randomized extended Gauss–Seidel and Kaczmarz methods. SIAM J. Matrix Anal. Appl. , 36(4):1590–1604, 2015

    2015

  30. [30]

    Faster randomized block Kaczmarz algorit hms

    Ion Necoara. Faster randomized block Kaczmarz algorit hms. SIAM J. Matrix Anal. Appl. , 40(4):1425– 1452, 2019

    2019

  31. [31]

    Randomized sketch desc ent methods for non-separable linearly con- strained optimization

    Ion Necoara and Martin Tak´ aˇ c. Randomized sketch desc ent methods for non-separable linearly con- strained optimization. IMA J. Numer. Anal. , 41(2):1056–1092, 2021

    2021

  32. [32]

    Randomized Kaczmarz solver for noisy l inear systems

    Deanna Needell. Randomized Kaczmarz solver for noisy l inear systems. BIT, 50(2):395–403, 2010

    2010

  33. [33]

    Paved with good intenti ons: analysis of a randomized block Kaczmarz method

    Deanna Needell and Joel A Tropp. Paved with good intenti ons: analysis of a randomized block Kaczmarz method. Linear Algebra Appl. , 441:199–221, 2014

    2014

  34. [34]

    Efficiency of coordinate descent methods on huge-scale optimization problems

    Yu Nesterov. Efficiency of coordinate descent methods on huge-scale optimization problems. SIAM J. Optim., 22(2):341–362, 2012

    2012

  35. [35]

    Introductory lectures on convex optimization: A basic cour se, volume 87

    Yurii Nesterov. Introductory lectures on convex optimization: A basic cour se, volume 87. Springer Sci- ence & Business Media, 2003

    2003

  36. [36]

    A method for solving the convex progra mming problem with convergence rate O(1 /k2)

    Yurii E Nesterov. A method for solving the convex progra mming problem with convergence rate O(1 /k2). In Dokl. akad. nauk Sssr , volume 269, pages 543–547, 1983

    1983

  37. [37]

    M-IH S: An accelerated randomized precondition- ing method avoiding costly matrix decompositions

    Ibrahim K Ozaslan, Mert Pilanci, and Orhan Arikan. M-IH S: An accelerated randomized precondition- ing method avoiding costly matrix decompositions. Linear Algebra Appl. , 678:57–91, 2023

    2023

  38. [38]

    LSQR: An alg orithm for sparse linear equations and sparse least squares

    Christopher C Paige and Michael A Saunders. LSQR: An alg orithm for sparse linear equations and sparse least squares. ACM Trans. Math. Softw. , 8(1):43–71, 1982

    1982

  39. [39]

    Some methods of speeding up the converge nce of iteration methods

    Boris T Polyak. Some methods of speeding up the converge nce of iteration methods. Comput. Math. Math. Phys. , 4(5):1–17, 1964

    1964

  40. [40]

    Extensions of block-projections met hods with relaxation parameters to inconsistent and rank-deficient least-squares problems

    Constantin Popa. Extensions of block-projections met hods with relaxation parameters to inconsistent and rank-deficient least-squares problems. BIT, 38(1):151–176, 1998

    1998

  41. [41]

    Characterization of the solutions se t of inconsistent least-squares problems by an extended Kaczmarz algorithm

    Constantin Popa. Characterization of the solutions se t of inconsistent least-squares problems by an extended Kaczmarz algorithm. Korean J. Comput. Appl. Math. , 6(1):51–64, 1999

    1999

  42. [42]

    An efficient ridge regression algorithm with pa rameter estimation for data analysis in ma- chine learning

    MP Rajan. An efficient ridge regression algorithm with pa rameter estimation for data analysis in ma- chine learning. SN Comput. Sci. , 3(2):171, 2022

    2022

  43. [43]

    Generalized Gearhart-Koshy accelera tion for the Kaczmarz method

    Janosch Rieger. Generalized Gearhart-Koshy accelera tion for the Kaczmarz method. Math. Comp. , 92(341):1251–1272, 2023

    2023

  44. [44]

    Sparse linear least -squares problems

    Jennifer Scott and Miroslav T ˚ uma. Sparse linear least -squares problems. Acta Numer. , 34:891–1010, 2025

    2025

  45. [45]

    A randomized Kacz marz algorithm with exponential conver- gence

    Thomas Strohmer and Roman Vershynin. A randomized Kacz marz algorithm with exponential conver- gence. J. Fourier Anal. Appl. , 15(2):262–278, 2009

    2009

  46. [46]

    On gree dy multi-step inertial randomized Kaczmarz method for solving linear systems

    Yansheng Su, Deren Han, Yun Zeng, and Jiaxin Xie. On gree dy multi-step inertial randomized Kaczmarz method for solving linear systems. Calcolo, 61(4):68, 2024

    2024

  47. [47]

    Connecting rand omized iterative methods with Krylov sub- spaces

    Yonghan Sun, Deren Han, and Jiaxin Xie. Connecting rand omized iterative methods with Krylov sub- spaces. arXiv preprint arXiv:2505.20602 , 2025

    work page arXiv 2025

  48. [48]

    Linear convergence of Gearhart-Koshy accelerated Kaczmarz methods for tensor linear systems

    Yijie Wang, Yonghan Sun, Deren Han, and Jiaxin Xie. Line ar convergence of Gearhart-Koshy acceler- ated Kaczmarz methods for tensor linear systems. arXiv preprint arXiv:2604.05816 , 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  49. [49]

    Coordinate descent algorithms

    Stephen J Wright. Coordinate descent algorithms. Math. Program., 151(1):3–34, 2015

    2015

  50. [50]

    Randomized itera tive methods for generalized absolute value equations: Solvability and error bounds

    Jiaxin Xie, Hou-Duo Qi, and Deren Han. Randomized itera tive methods for generalized absolute value equations: Solvability and error bounds. SIAM J. Optim. , 35(3):1731–1760, 2025

    2025

  51. [51]

    Local ridge regre ssion for face recognition

    Hui Xue, Yulian Zhu, and Songcan Chen. Local ridge regre ssion for face recognition. Neurocomputing, 72(4-6):1342–1346, 2009

    2009

  52. [52]

    On adap tive stochastic heavy ball momentum for solving linear systems

    Yun Zeng, Deren Han, Yansheng Su, and Jiaxin Xie. On adap tive stochastic heavy ball momentum for solving linear systems. SIAM J. Matrix Anal. Appl. , 45(3):1259–1286, 2024

    2024

  53. [53]

    On adap tive stochastic extended iterative methods for solving least squares

    Yun Zeng, Deren Han, Yansheng Su, and Jiaxin Xie. On adap tive stochastic extended iterative methods for solving least squares. Math. Comp. , DOI: https://doi.org/10.1090/mcom/4168, 2025. 34 YUN ZENG, JIAN-FENG CAI, DEREN HAN, AND JIAXIN XIE

    work page doi:10.1090/mcom/4168 2025

  54. [54]

    Stocha stic dual coordinate descent with adaptive heavy ball momentum for linearly constrained convex optimi zation

    Yun Zeng, Deren Han, Yansheng Su, and Jiaxin Xie. Stocha stic dual coordinate descent with adaptive heavy ball momentum for linearly constrained convex optimi zation. Numer. Math. , 158(2):749–794, 2026

    2026

  55. [55]

    Anastasios Zouzias and Nikolaos M. Freris. Randomized extended Kaczmarz for solving least squares. SIAM J. Matrix Anal. Appl. , 34(2):773–793, 2013. School of Mathematical Sciences, Beihang University, Beij ing, 100191, China. Email address : zengyun@buaa.edu.cn Department of Mathematics, Hong Kong University of Science and Technology, Clear W ater Bay, K...

    2013