Pith. sign in

REVIEW 2 major objections 5 minor 51 references

A Globalized Semismooth Newton Method for Prox-regular Optimization Problems

T0 review · 2 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A Newton method that works on nonconvex sparse optimization

desk verdict A solid new globalized semismooth Newton method for prox-regular composites, but the key well-definedness lemma has a proof gap at empty-subdifferential points that hits the paper's own ℓ_q example. read the letter →

arxiv 2509.05765 v1 pith:5JQBV4AA submitted 2025-09-06 math.OC

Yuqia Wu , Pengcheng Wu , Yaohua Hu , Shaohua Pan , Xiaoqi Yang This is my paper · ORCID
classification math.OC MSC 90C2690C5365K0549J52
keywords semismoothNewtonmethodprox-regularfunctionsnonconvexcompositeoptimizationKurdyka-Łojasiewiczinequalitymetricsubregularitysparseregularizationforward-backwardenvelope
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper designs an algorithm that alternates a safeguarded proximal-gradient step with a semismooth Newton step for composite problems where the smooth part is twice differentiable and the nonsmooth part is nonconvex but prox-regular. The central claim is that the proximal residual of such a nonsmooth term becomes single-valued and locally Lipschitz near any point, so a Newton system with a well-defined generalized Jacobian can be formed. Under a Kurdyka-Łojasiewicz exponent condition, the whole sequence of iterates converges to an L-stationary point; under an additional metric-subregularity condition, the distance to the stationary set shrinks superlinearly, without requiring the limit point to be isolated. This would be the first globalized semismooth Newton method with both global and superlinear convergence of the iterates for nonconvex prox-regular problems, and the paper demonstrates it on ℓq and fused zero-norm sparse models.

What carries the argument

The central object is the proximal residual Rγ(x)=γ⁻¹(x−Pγg(x−γ∇f(x))), together with the generalized Jacobian surrogate ∂²Fγ(x)=γ⁻¹Qγ(x)(I−W Qγ(x)) with Qγ(x)=I−γ∇²f(x). Proposition 2.1 supplies the single-valuedness and local Lipschitz continuity of the proximal mapping for prox-regular functions, which is what makes Rγ a well-defined semismooth system and gives a computable Clarke-type Jacobian. The algorithm's hybrid structure—a proximal-gradient safeguard step followed by a semismooth Newton step with line search on the forward-backward envelope—is what carries the global-to-local convergence argument.

What would settle it

Construct a prox-regular, prox-bounded g satisfying Assumptions 1–3 for which F is a KL function with exponent below 1/2 and run Algorithm 1 from an initial point where step (2a) must backtrack: if a finite admissible stepsize exists but the iterates fail to converge to an L-stationary point, the KL-exponent range in Theorem 4.1 is violated. Alternatively, take a fused zero-norm problem with a large stationary set X* and check whether dist(yk, X*) still vanishes superlinearly; if Assumption 4(ii) (metric subregularity of the residual) fails, the superlinear claim should cease to hold.

Watch

Extended reading notes

Core claim

The paper proves (Proposition 2.1) that for a proper, lower semicontinuous, prox-bounded function h that is prox-regular at x, the proximal mapping Pγh is single-valued, locally Lipschitz continuous, and monotone on a neighborhood of x for all sufficiently small γ. This upgrades the classical local result, which is centered at x+γv, to a neighborhood of x independent of v. With this, the composite residual Rγ(x)=γ⁻¹(x−Tγ(x)), where Tγ(x)=Pγg(x−γ∇f(x)), becomes a locally Lipschitz nonsmooth system whose Clarke Jacobian exists, so semismooth Newton can be applied. Algorithm 1 backtracks to pick a stepsize γ̃k in this admissible range, takes a proximal-gradient step to a feasible point xk, then

Load-bearing premise

The algorithm requires g to be prox-regular on its whole effective domain and prox-bounded with threshold at least 1/L; if this global regularity fails, the backtracking step may never find a stepsize at which the residual is single-valued and locally Lipschitz, and the method may not be well-defined.

Editorial extensions

If this is right

  • If Theorem 4.1 is correct, this is the first globalized semismooth Newton method whose whole iterate sequence is proven to converge for nonconvex prox-regular g under only a KL-exponent assumption, without strong prox-regularity or convexity of g.
  • The superlinear rate in Theorem 5.1 holds for nonisolated L-stationary points, measured by distance to the entire stationary set X*, not just to a single limit point.
  • The single-valuedness result gives a practical algorithmic criterion: once the proximal mapping is single-valued at a point, it is automatically locally Lipschitz there, which is exactly what the backtracking step (2a) checks.
  • The method is directly applicable to ℓq quasi-norm and fused zero-norm regularized problems, where the numerical experiments show comparable or better objective values and running times than two existing second-order methods.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The single-valuedness check developed in the appendix for fused zero-norm proximal mappings could be reused independently as an acceptance test for Newton-type steps in other proximal solvers, not only within this algorithm.
  • The superlinear result suggests that sparsity-inducing problems with a continuum of stationary points—previously treated mostly by first-order analysis—may admit fast second-order methods; this can be tested by running the algorithm on fused zero-norm instances deliberately engineered with nonunique solutions.
  • A natural next step, left implicit in the paper, is whether the KL-exponent interval [1/2, 1/(2−ϱ)] can be widened or replaced by a sharpness condition that is easier to verify for specific regularizers such as the ℓq quasi-norm.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

2 major / 5 minor

Summary. The paper studies the composite optimization problem min f(x)+g(x), where f is twice differentiable with Lipschitz gradient/Hessian and g is prox-regular and prox-bounded. The main contribution is a globalization of the semismooth Newton method by interleaving proximal-gradient (PG) steps with semismooth Newton steps. The proximal residual R_γ is used as the nonsmooth equation; a key preliminary result (Proposition 2.1) asserts that the proximal mapping of a prox-regular function is locally single-valued, Lipschitz, and monotone, which is then used to ensure that the Newton step is well defined. The authors prove full-sequence convergence to an L-stationary point under a KL exponent condition θ∈[1/2,1/(2−ϱ)] (Theorem 4.1), and superlinear convergence of dist(y_k,X*) to 0 under an additional metric subregularity assumption and conditions (C1)–(C3) (Theorem 5.1). Numerical experiments on ℓ_q regularized problems and fused zero-norm regularized problems compare favorably with existing second-order methods.

Significance. If the results are correct as stated, this is a substantial advance: it would be the first globalized semismooth Newton method for nonconvex prox-regular g with both full-sequence convergence and superlinear convergence to possibly nonisolated stationary points. The paper contains detailed proofs, carefully tracks the KL exponent range, and includes useful conditions for the superlinear rate. The preliminary numerical results suggest practical competitiveness. The main limitation is that the theory and the numerical examples are not fully aligned: the examples include functions, such as ℓ_q quasi-norms and zero-norm regularizers, whose limiting subdifferential may be empty at the sparse points that are the intended solutions, while the proof of the key proximal-mapping result appears to require a nonempty subdifferential.

major comments (2)
  1. [§6.2, Proposition 2.1, Assumption 2] The proof of Proposition 2.1 begins with r := 1 + 4 min_{v∈∂h(\bar x)} ||v||. This expression is undefined when ∂h(\bar x)=∅. Under Definition 2.2, prox-regularity at \bar x is a universal statement over v∈∂h(\bar x), so it is vacuously true when the limiting subdifferential is empty. For g=λ||x||_q^q with 0<q<1, ∂g(0)=∅: coordinate derivatives diverge as any coordinate approaches 0. Sparse solutions of ℓ_q problems have zero coordinates, so this is exactly the situation arising in the experiments of Section 6.2. Consequently either (a) prox-regularity should be read as implicitly requiring a nonempty subdifferential, in which case Assumption 2 excludes the ℓ_q example and Theorem 4.1 does not cover it; or (b) empty-subdifferential points are admitted, in which case Proposition 2.1 is unproved and Lemma 3.1(ii) has no certified step (2a) for such iterates. The issue is load-bearing becau
  2. [§1.1, Assumption 1(ii), §6.3] Assumption 1(ii) requires g to be proper and lower semicontinuous. The standard zero-norm function ||x||_0 (0 at x=0 and 1 otherwise) is not lsc at x=0: any neighborhood of 0 contains points with ||x||_0>0, so liminf at 0 is at least 1 while the value is 0. The same applies to the fused zero-norm function g_1=λ0||B·||0+λ||·||0+δ_Ω used in Section 6.3 when 0∈int Ω. Thus, unless a different lsc surrogate or a nonstandard definition is intended, the numerical examples do not satisfy the standing assumptions of the convergence theory. If the authors intend to include zero-norm regularizers, they should state the exact definition and explain how Assumption 1(ii) and prox-regularity are satisfied; otherwise these experiments are outside the hypotheses of Theorems 4.1 and 5.1.
minor comments (5)
  1. [Algorithm 1 input, §3] The input line contains a duplicated inequality: '0< ς < ς' should presumably be '0< ς < \bar ς', matching the notation σ_k∈[σ,\bar σ] and ς_k∈[ς,\bar ς] used later.
  2. [Definition 2.2] In the displayed definition of prox-regularity, the condition 'h(x)<h(x)+ε' appears to contain a typo: the second h(x) should be h(\bar x), so that the inequality is h(x)<h(\bar x)+ε.
  3. [§6.1] Typo: 'nonnagative integer' should be 'nonnegative integer'.
  4. [§6.3, Table 3] The PSNR formula is reported as 10 log10(n / ||x−x*||). Standard PSNR uses ||x−x*||^2 in the denominator. Please clarify which definition is used.
  5. [§2.1, proof of Proposition 2.1] The notation r:=1+4 min_{v∈∂h(\bar x)} ||v|| is used before the empty-subdifferential issue is resolved; even if the empty case is excluded, the reader would benefit from an explicit statement that ∂h(\bar x) is assumed nonempty in Proposition 2.1.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; convergence theorems are derived from stated assumptions with self-citations only as auxiliary published tools.

full rationale

The derivation chain is not circular. Algorithm 1's well-definedness (Lemma 3.1) is proved from Lemma 2.2(ii), Proposition 2.2, and the descent property of F_{eγ_k}; none of these assume the convergence conclusion. Global convergence (Theorem 4.1) takes the KL property of F as a regularity premise and derives summability of successive differences through the standard KL/descent argument; the KL property is not obtained from the iterates. The superlinear result (Theorem 5.1) adds Assumption 4 and conditions (C1)-(C3) and derives the superlinear direction property from them, rather than importing the conclusion. Citations to the authors' prior work ([43,44,45]) supply KL-exponent examples, proximal-map algorithms, and the L-stationary/critical equivalence; these are auxiliary published tools, not load-bearing assumptions of the central convergence theorem. Other references ([42,48,39,9]) are independent. The numerical section is benchmarking and does not fit a parameter that is later renamed a prediction. No equation in the proof reduces by construction to its own input, and no uniqueness theorem is imported from the authors to force a choice. One correctness gap exists but is not circularity: in Proposition 2.1 the proof begins 'Let r:= 1+4 min_{v in ∂h(x)} ||v||', which is undefined when ∂h(x)=∅; this can affect applicability to ℓ_q at zero coordinates and is a rigor/applicability concern, not a self-referential derivation.

Assumptions & free parameters 0 free parameters · 6 assumptions · 0 invented entities

The central claim rests on the prox-regularity and prox-boundedness of the nonsmooth term, the KL exponent range, and standard variational analysis tools. No new physical or mathematical entities are postulated; the algorithm parameters are not fitted to data.

assumptions (6)
  • domain assumption f is twice differentiable on open O⊃dom g, with ∇f and ∇^2f Lipschitz; g proper lsc with computable proximal mapping; F lower bounded (Assumption 1)
    Standard composite optimization setting; used throughout.
  • domain assumption g is prox-regular over dom g and prox-bounded with threshold γ_g ≥ 1/L (Assumption 2)
    Needed for Proposition 2.1/2.2 to guarantee existence of eγ_k with single-valued T_eγ_k in step (2a).
  • domain assumption The iterate sequence {x_k} is bounded (Assumption 3)
    Ensures compactness for Lemma 3.2; can be replaced by level-boundedness of F.
  • domain assumption F is a KL function with exponent θ∈[1/2, 1/(2-ϱ)] (Theorem 4.1)
    Core assumption for global convergence of the whole sequence; stronger than plain KL, but holds for many sparsity functions.
  • domain assumption Superlinear result: x* local min of F; R* metrically subregular; F* constant on X* near x*; conditions C1-C3 (Assumption 4 and C1-C3)
    Mild but not fully verified for the examples; C2 (positive semidefiniteness of ∂^2F* on X*) is especially strong.
  • standard math Use of external results: Rockafellar-Wets [39] theorems on prox-regular functions, KL calculus [48], Clarke Jacobian calculus [13,11]
    Background from variational analysis and nonsmooth analysis.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Globalized Semismooth Newton Method for Prox-regular Optimization Problems." pith.science (2026). https://pith.science/paper/5JQBV4AA

@misc{pith2026250905765,
  author       = {Pith},
  title        = {Pith review of: A Globalized Semismooth Newton Method for Prox-regular Optimization Problems},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5JQBV4AA}},
  note         = {Machine review of arXiv:2509.05765}
}
abstract

We are concerned with a class of nonconvex and nonsmooth composite optimization problems, comprising a twice differentiable function and a prox-regular function. We establish a sufficient condition for the proximal mapping of a prox-regular function to be single-valued and locally Lipschitz continuous. By virtue of this property, we propose a hybrid of proximal gradient and semismooth Newton methods for solving these composite optimization problems, which is a globalized semismooth Newton method. The whole sequence is shown to converge to an $L$-stationary point under a Kurdyka-{\L}ojasiewicz exponent assumption. Under an additional error bound condition and some other mild conditions, we prove that the sequence converges to a nonisolated $L$-stationary point at a superlinear convergence rate. Numerical comparison with several existing second order methods reveal that our approach performs comparably well in solving both the $\ell_q(0<q<1)$ quasi-norm regularized problems and the fused zero-norm regularization problems.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

51 extracted references · 49 canonical work pages

  1. [1]

    Ahookhosh, A

    M. Ahookhosh, A. Themelis, and P. Patrinos , A B regman forward-backward linesearch algorithm for nonconvex composite optimization: superlinear convergence to nonisolated local minima , SIAM Journal on Optimization, 31 (2021), pp. 653--685

    work page 2021

  2. [2]

    F. J. A. Artacho and M. H. Geoffroy , Metric subregularity of the convex subdifferential in banach spaces , arXiv preprint arXiv:1303.3654, (2013)

    work page Pith review arXiv 2013

  3. [3]

    Attouch and J

    H. Attouch and J. Bolte , On the convergence of the proximal algorithm for nonsmooth functions involving analytic features , Mathematical Programming, 116 (2009), pp. 5--16

    work page 2009

  4. [4]

    Attouch, J

    H. Attouch, J. Bolte, P. Redont, and A. Soubeyran , Proximal alternating minimization and projection methods for nonconvex problems: An approach based on the K urdyka- ojasiewicz inequality , Mathematics of Operations Research, 35 (2010), pp. 438--457

    work page 2010

  5. [5]

    Attouch, J

    H. Attouch, J. Bolte, and B. F. Svaiter , Convergence of descent methods for semi-algebraic and tame problems: proximal algorithms, forward--backward splitting, and regularized G auss- S eidel methods , Mathematical Programming, 137 (2013), pp. 91--129

    work page 2013

  6. [6]

    Bareilles, F

    G. Bareilles, F. Iutzeler, and J. Malick , Newton acceleration on manifolds identified by proximal gradient methods , Mathematical Programming, (2022), pp. 1--34

    work page 2022

  7. [7]

    D. P. Bertsekas , Projected N ewton methods for optimization problems with simple constraints , SIAM Journal on control and Optimization, 20 (1982), pp. 221--246

    work page 1982

  8. [8]

    Bhatia , Matrix analysis , vol

    R. Bhatia , Matrix analysis , vol. 169, Springer Science & Business Media, 2013

    work page 2013

Show all 51 references
  1. [9]

    Bolte, S

    J. Bolte, S. Sabach, and M. Teboulle , Proximal alternating linearized minimization for nonconvex and nonsmooth problems , Mathematical Programming, 146 (2014), pp. 459--494

    2014

  2. [10]

    W. Cao, J. Sun, and Z. Xu , Fast image deconvolution using closed-form thresholding formulas of L _q(q= 1 2 , 2 3 ) regularization , Journal of Visual Communication and Image representation, 24 (2013), pp. 31--41

    2013

  3. [11]

    Z. X. Chan and D. Sun , Constraint nondegeneracy, strong regularity, and nonsingularity in semidefinite programming , SIAM Journal on optimization, 19 (2008), pp. 370--396

    2008

  4. [12]

    Charisopoulos and D

    V. Charisopoulos and D. Davis , A superlinearly convergent subgradient method for sharp semismooth problems , Mathematics of Operations Research, 49 (2024), pp. 1678--1709

    2024

  5. [13]

    F. H. Clarke , Optimization and nonsmooth analysis , SIAM, 1990

    1990

  6. [14]

    De Marchi and A

    A. De Marchi and A. Themelis , Proximal gradient algorithms under local L ipschitz gradient continuity: A convergence and robustness analysis of PANOC , Journal of Optimization Theory and Applications, 194 (2022), pp. 771--794

    2022

  7. [15]

    Fan and R

    J. Fan and R. Li , Variable selection via nonconcave penalized likelihood and its oracle properties , Journal of the American statistical Association, 96 (2001), pp. 1348--1360

    2001

  8. [16]

    Gfrerer , On a globally convergent semismooth* N ewton method in nonsmooth nonconvex optimization , Computational Optimization and Applications, (2025), pp

    H. Gfrerer , On a globally convergent semismooth* N ewton method in nonsmooth nonconvex optimization , Computational Optimization and Applications, (2025), pp. 1--58

    2025

  9. [17]

    Gfrerer and J

    H. Gfrerer and J. V. Outrata , On a semismooth* N ewton method for solving generalized equations , SIAM Journal on Optimization, 31 (2021), pp. 489--517

    2021

  10. [18]

    Hiriart-Urruty, J.-J

    J.-B. Hiriart-Urruty, J.-J. Strodiot, and V. H. Nguyen , Generalized H essian matrix and second-order optimality conditions for problems with C _ 1,1 data , Applied mathematics and optimization, 11 (1984), pp. 43--56

    1984

  11. [19]

    J. Hu, K. Deng, J. Wu, and Q. Li , A projected semismooth N ewton method for a class of nonconvex composite programs with strong prox-regularity , Journal of Machine Learning Research, 25 (2024), pp. 1--32

    2024

  12. [20]

    J. Hu, T. Tian, S. Pan, and Z. Wen , On the local convergence of the semismooth N ewton method for composite optimization , arXiv preprint arXiv:2211.01127, (2022)

    2022 arXiv

  13. [21]

    Huang, J

    L. Huang, J. Jia, B. Yu, B.-G. Chun, P. Maniatis, and M. Naik , Predicting execution time of computer programs using sparse polynomial regression , Advances in Neural Information Processing Systems, 23 (2010), pp. 883--891

    2010

  14. [22]

    Kanzow and T

    C. Kanzow and T. Lechner , Globalized inexact proximal N ewton-type methods for nonconvex composite functions , Computational Optimization and Applications, 78 (2021), pp. 377--410

    2021

  15. [23]

    P. D. Khanh, B. S. Mordukhovich, and V. T. Phat , Coderivative-based N ewton methods in structured nonconvex and nonsmooth optimization , arXiv preprint arXiv:2403.04262, (2024)

    2024

  16. [24]

    P. D. Khanh, B. S. Mordukhovich, V. T. Phat, and D. B. Tran , Globally convergent coderivative-based generalized N ewton methods in nonsmooth optimization , Mathematical Programming, (2023), pp. 1--57

    2023

  17. [25]

    J. D. Lee, Y. Sun, and M. A. Saunders , Proximal N ewton-type methods for minimizing composite functions , SIAM Journal on Optimization, 24 (2014), pp. 1420--1443

    2014

  18. [26]

    Li and T

    G. Li and T. K. Pong , Calculus of the exponent of K urdyka-- ojasiewicz inequality and its applications to linear convergence of first-order methods , Foundations of Computational Mathematics, 18 (2018), pp. 1199--1232

    2018

  19. [27]

    Liebscher and G

    V. Liebscher and G. Winkler , A potts model for segmentation and jump-detection , in Proceedings S4G International Conference on Stereology, Spatial Statistics and Stochastic Geometry, Prague June, vol. 21, Citeseer, 1999, pp. 185--190

    1999

  20. [28]

    R. Liu, S. Pan, Y. Wu, and X. Yang , An inexact regularized proximal N ewton method for nonconvex and nonsmooth optimization , Computational Optimization and Applications, 88 (2024), pp. 603--641

    2024

  21. [29]

    Luo and P

    Z.-Q. Luo and P. Tseng , Error bounds and convergence analysis of feasible descent methods: a general approach , Annals of Operations Research, 46 (1993), pp. 157--178

    1993

  22. [30]

    Milzarek and M

    A. Milzarek and M. Ulbrich , A semismooth N ewton method with multidimensional filter globalization for l_1 -optimization , SIAM Journal on Optimization, 24 (2014), pp. 298--333

    2014

  23. [31]

    B. S. Mordukhovich, X. Yuan, S. Zeng, and J. Zhang , A globally convergent proximal N ewton-type method in nonsmooth convex optimization , Mathematical Programming, 198 (2023), pp. 899--936

    2023

  24. [32]

    Ochs , Local convergence of the heavy-ball method and i P iano for non-convex optimization , Journal of Optimization Theory and Applications, 177 (2018), pp

    P. Ochs , Local convergence of the heavy-ball method and i P iano for non-convex optimization , Journal of Optimization Theory and Applications, 177 (2018), pp. 153--180

    2018

  25. [33]

    Ouyang and A

    W. Ouyang and A. Milzarek , A trust region-type normal map-based semismooth N ewton method for nonsmooth nonconvex composite optimization , Mathematical Programming, (2024), pp. 1--47

    2024

  26. [34]

    Pang and L

    J.-S. Pang and L. Qi , Nonsmooth equations: motivation and algorithms , SIAM Journal on optimization, 3 (1993), pp. 443--465

    1993

  27. [35]

    Patrinos and A

    P. Patrinos and A. Bemporad , Proximal N ewton methods for convex composite optimization , in 52nd IEEE Conference on Decision and Control, IEEE, 2013, pp. 2358--2363

    2013

  28. [36]

    Poliquin and R

    R. Poliquin and R. Rockafellar , Prox-regular functions in variational analysis , Transactions of the American Mathematical Society, 348 (1996), pp. 1805--1838

    1996

  29. [37]

    Qi and J

    L. Qi and J. Sun , A nonsmooth version of N ewton's method , Mathematical programming, 58 (1993), pp. 353--367

    1993

  30. [38]

    S. M. Robinson , Some continuity properties of polyhedral multifunctions , Springer, 1981

    1981

  31. [39]

    R. T. Rockafellar and R. J.-B. Wets , Variational analysis , vol. 317, Springer Science & Business Media, 2009

    2009

  32. [40]

    Stella, A

    L. Stella, A. Themelis, and P. Patrinos , Forward--backward quasi- N ewton methods for nonsmooth optimization problems , Computational Optimization and Applications, 67 (2017), pp. 443--487

    2017

  33. [41]

    Themelis, M

    A. Themelis, M. Ahookhosh, and P. Patrinos , On the acceleration of forward-backward splitting via an inexact N ewton method , in Splitting Algorithms, Modern Operator Theory, and Applications, Springer, 2019, pp. 363--412

    2019

  34. [42]

    Themelis, L

    A. Themelis, L. Stella, and P. Patrinos , Forward-backward envelope for the sum of two nonconvex functions: Further properties and nonmonotone linesearch algorithms , SIAM Journal on Optimization, 28 (2018), pp. 2274--2303

    2018

  35. [43]

    Y. Wu, S. Pan, and S. Bi , Kurdyka-- ojasiewicz property of zero-norm composite functions , Journal of Optimization Theory and Applications, 188 (2021), pp. 94--112

    2021

  36. [44]

    Y. Wu, S. Pan, and X. Yang , A regularized N ewton method for _q -norm composite optimization problems , SIAM Journal on Optimization, 33 (2023), pp. 1676--1706

    2023

  37. [45]

    height 2pt depth -1.6pt width 23pt, An inexact projected regularized N ewton method for fused zero-norms regularization problems , Journal of Machine Learning Research, 25 (2024), pp. 1--48

    2024

  38. [46]

    X. Xiao, Y. Li, Z. Wen, and L. Zhang , A regularized semi-smooth N ewton method with projection steps for composite convex programs , Journal of Scientific Computing, 76 (2018), pp. 364--389

    2018

  39. [47]

    Z. Xu, X. Chang, F. Xu, and H. Zhang , _ 1/2 regularization: A thresholding representation theory and a fast solver , IEEE Transactions on Neural Networks and Learning Systems, 23 (2012), pp. 1013--1027

    2012

  40. [48]

    P. Yu, G. Li, and T. K. Pong , Kurdyka-- L ojasiewicz exponent via inf-projection , Foundations of Computational Mathematics, 22 (2022), pp. 1171--1217

    2022

  41. [49]

    M.-C. Yue, Z. Zhou, and A. M.-C. So , A family of inexact SQA methods for non-smooth convex minimization with provable convergence guarantees based on the L uo- T seng error bound property , Mathematical Programming, 174 (2019), pp. 327--358

    2019

  42. [50]

    Zhang , Nearly unbiased variable selection under minimax concave penalty , Annals of Statistics, (2010), pp

    C.-H. Zhang , Nearly unbiased variable selection under minimax concave penalty , Annals of Statistics, (2010), pp. 894--942

    2010

  43. [51]

    S. Zhou, X. Xiu, Y. Wang, and D. Peng , Revisiting l_q (0 q< 1) norm regularized optimization , arXiv preprint arXiv:2306.14394, (2023)

    2023 arXiv

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.