betaFeedback

WorksMathematics

Quasi-norms on Rd\mathbb{R}^d: Characterization and Approximation by Conic Splines

Abstract

We first show that a regularizer on Rd\mathbb{R}^d that is continuous, symmetric, and amplitude-equivariant is of the form g=qγg=q^γ with γ>0γ>0, where qq is continuous and absolutely homogeneous; we call the set of such qq the CH class, whose convex members are the seminorms. Within the CH class, we single out two configurations of interest for regularization: the quasi-norms, which are exactly the positive-definite members of the family, and the quasi-seminorms with null space control, which are the composition of a quasi-norm with a linear map that sets the null space. We then propose a universal parametrization of the class by conic splines, the continuous functions that are linear on each cone of a polyhedral fan. Specifically, we show that the polyhedral CH functionals are exactly the symmetric non-negative conic splines, and that these admit a stable nodal representation in a basis of conic B-splines. Finally, we determine the rates of approximation: the conic spline interpolant converges at the second-order rate h2≍n−2/(d−1)h^2\asymp n^{-2/(d-1)}, in the L∞L_\infty-norm on the hyper-sphere when the angular profile is twice differentiable, and in the L1L_1-norm when its spherical Hessian variation is bounded, which covers the polyhedral members of the class. The rate is sharp, already for the Euclidean norm. The planar case is worked out in full, with the trigonometric hat functions as basis and an ℓ1\ell_1-norm of second differences as regularity index.

Introduction

The functionals that this paper studies arise from the learning of a regularizer. The reconstruction of a signal s∈Rd\mathbf{s} \in\mathbb{R}^{d} from the noisy linear measurements y=Hs+ϵ∈RM\mathbf{y} = \mathbf{H}\mathbf{s} + \boldsymbol{\epsilon} \in\mathbb{R}^{M} is formulated as

s⋆=arg min⁡s∈Rd 12∥y−Hs∥22+λgθ(s),(1)\mathbf{s}^{\star} = \underset{\mathbf{s}\in\mathbb{R}^{d}}{\operatorname{arg\,min}}\ \frac{1}{2}\left\lVert\mathbf{y}-\mathbf{H}\mathbf{s}\right\rVert_{2}^{2}+\lambda g_{\theta}(\mathbf{s}), \tag*{(1)}

where the regularizer gθg_{\theta} carries the prior information and where its parameters θ\theta are trained on representative data [10, 40, 7, 41]. In practice, the quality of the reconstruction is tuned through a proper adjustment of the regularization strength λ>0\lambda> 0. The appeal of (1)—as opposed to a direct regression by a neural network—is that the reconstruction remains the solution of an optimization problem, so that the existence of a solution, the convergence of the iterative scheme, and the stability of the map y↦s⋆\mathbf{y} \mapsto\mathbf{s}^{\star} remain within reach; but those statements are only as strong as the structural constraints that one is willing to impose on gθg_{\theta}. Three requirements are natural: amplitude equivariance, which asks that the amplification of the data by a factor α≠0\alpha\ne0 amplify the reconstruction by the same factor up to a readjustment of λ\lambda; symmetry, gθ(−s)=gθ(s)g_{\theta}(-\mathbf{s})=g_{\theta}(\mathbf{s}), since the sign of a contrast carries no prior preference; and continuity, the bare minimum for any statement about convergence. Under these three requirements together with convexity, the admissible regularizers collapse to the powers g=pγg=p^{\gamma}, γ≥1\gamma\ge1, of a seminorm pp [57 Theorem 2.3]. The same paper also provides a universal parametrization of such functionals by means of learnable polyhedral norms and a demonstration of usage for biomedical image reconstruction.

Objectives and Contributions

The purpose of the present study is to extend the above framework by dropping the convexity constraint, which the sparsity-promoting penalties of practical interest—foremost among them the ℓp\ell_{p} “norms” with 0<p<10 < p < 1—do not satisfy. This raises three questions, which organize the paper: the identification of the class of relevant regularization functionals; the formulation of a parametric framework that is universal for the class; and the rates of approximation that the parametrization achieves.

1. Identification and categorization of the class. We first show that the continuous, symmetric, and amplitude-equivariant functionals are exactly the powers g=qγg=q^{\gamma}, γ>0\gamma> 0, of the functionals qq that are continuous and absolutely homogeneous,

∀x∈Rd, ∀α∈R:q(αx)=∣α∣q(x),(2)\forall x \in\mathbb{R}^{d},\ \forall\alpha\in\mathbb{R}: \quad q(\alpha x)=|\alpha|q(x), \tag*{(2)}

which we refer to as the CH class; in particular, gg is convex if and only if γ≥1\gamma\ge1 and qq is a seminorm (Proposition 2.2). Since the exponent γ\gamma is immaterial for (1), the object to be learned is qq itself. Within the CH class on Rd\mathbb{R}^{d}, the quasi-norms are exactly the positive-definite members (Theorem 2.4). Likewise, the quasi-seminorms are exactly the members that are comparable to a seminorm; their null space Nq\mathcal{N}_{q} is then a linear subspace, and qq is comparable to the distance to it (Theorem 2.5). The null space has an essential role in regularization theory: it is the mechanism by which one favors a type of solution—the constant signals, say, when Nq\mathcal{N}_{q} is the kernel of a finite-difference operator—while the data alone decide the component of the reconstruction along Nq\mathcal{N}_{q}. In our framework, this type of regularizer is implemented as the composition qˉ∘P\bar{q}\circ\mathbf{P} of a quasi-norm qˉ\bar{q} with a matrix P\mathbf{P} that sets the null space. Theorem 2.7 characterizes these regularizers as the quasi-seminorms that are invariant along their null space. On the geometric side, the continuous quasi-norms are the gauges of the symmetric star-shaped bodies with a continuous positive radial function (Theorem 2.8).

Universal parametrization by conic splines

A parametrization of the class should be: (i) universal, in that it covers the class in the limit of infinitely many parameters; (ii) stable, so that the parameters can be adjusted by a descent algorithm; and (iii) free of constraints, so that the admissibility of the learned functional is guaranteed by construction, as the polyhedral norms achieve in the convex theory.

To this end, we introduce the conic splines: the continuous functions that are linear on each cone of a complete simplicial fan of Rd\mathbb{R}^{d} or, equivalently, the 1-homogeneous extensions of the piecewise-linear functions on a triangulation of the hyper-sphere. We show that the symmetric non-negative conic splines are precisely the polyhedral members of the CH class (Theorem 3.3), and that the conic B-splines—the conic analogues of the Courant hat functions—form a nodal basis of the conic splines on a fan, with the values of the functional along the rays as coefficients and with explicit Riesz bounds (Proposition 3.2). The parametrization of a polyhedral CH functional by its nodal values is thus free of constraints up to the non-negativity and the symmetry of those values. By Theorem 2.4, the parametrized functional is a polyhedral quasi-norm if and only if all the nodal values are (strictly) positive. The composition of the latter with a matrix P∈Rd′×d\mathbf{P}\in\mathbb{R}^{d'\times d} then yields a constraint-free parametrization of the quasi-seminorms with the prescribed null space Nq=ker⁡(P)\mathcal{N}_q = \ker(\mathbf{P}).

3. Rates of approximation and their sharpness. Universality is a qualitative statement, which, for approximation purposes, needs to be complemented by a rate: how many rays a prescribed accuracy costs, which regularity of the profile this depends on, and whether the rate is the best possible.

For a 11-homogeneous qq that is C2C^2 away from the origin, we show that the conic spline interpolant qn=∑iq(ri)φiq_n = \sum_i q(\mathbf{r}_i)\varphi_i on a shape-regular fan with rays r1,…,rn\mathbf{r}_1,\ldots,\mathbf{r}_n and maximal spherical cell diameter hnh_n satisfies

∥q−qn∥L∞(Sd−1)≤Chn2HLip⁡Sd−1(q),\lVert q-q_n\rVert_{L_\infty(\mathbb{S}^{d-1})} \le C h_n^2 \operatorname{HLip}_{\mathbb{S}^{d-1}}(q),

where HLip⁡Sd−1(q)\operatorname{HLip}_{\mathbb{S}^{d-1}}(q) is the supremum of the spectral norm of the Hessian of qq over the hyper-sphere; for quasi-uniform rays, hn≍n−1/(d−1)h_n \asymp n^{-1/(d-1)} and the rate is n−2/(d−1)n^{-2/(d-1)} (Theorem 4.4). The rate is sharp: by Gruber’s lower bound on the approximation of a convex body by an inscribed polytope with nn vertices [25], no conic spline on a fan with at most nn rays approximates the Euclidean norm better than cdn−2/(d−1)c_d n^{-2/(d-1)}, so that the approximation saturates at the second order however smooth the quasi-norm is (Theorem 4.4, Item 3). We then extend our second-order estimate to the CH functionals with a bounded spherical Hessian variation HTV⁡Sd−1(q)\operatorname{HTV}_{\mathbb{S}^{d-1}}(q), a quantity defined by duality that is finite for the polyhedral members of the class: a preaveraged conic spline then converges in L1(Sd−1)L_1(\mathbb{S}^{d-1}) at the rate hn2HTV⁡Sd−1(q)h_n^2 \operatorname{HTV}_{\mathbb{S}^{d-1}}(q) (Theorem 4.7). In the planar case, the conic B-splines on a uniform polar grid are trigonometric hat functions and the spherical Hessian variation of a polyhedral profile is an ℓ1\ell_1-norm of a second difference of the nodal values (§ 6), which is the penalty of the learning problem.

Related Works

Non-convex sparsity-promoting penalties have a long history in signal reconstruction: the ℓp\ell_p quasi-norms with 0<p<10 < p < 1 recover sparse signals under weaker conditions than the ℓ1\ell_1-norm [13, 17, 18], the reweighted ℓ1\ell_1 schemes approximate them [12], and the non-convex penalties of statistics and image processing [16, 59, 43, 44, 54, 52, 34] trade convexity for a better recovery of edges and of sparse supports. The learning of a regularizer from data is now a field of its own [7, 41, 26]. Recent contributions include convex architectures [5, 22] and their weakly convex extensions [23, 53], adversarial and network-based regularizers [38, 37, 31], learned priors and bilevel schemes [47, 33, 14], and the plug-and-play methods that replace the regularizer by a learned denoiser [58, 46, 49, 28, 30, 27]. The role of amplitude equivariance in this context, and the universality of the polyhedral norms for the convex regularizers, are established in [57], of which the present paper is the non-convex counterpart; the gauges of star bodies also appear as the natural regularizers in the critic-based learning of [35, 36]. The consequences of the homogeneity of qq for the variational problem (1)—existence of a minimizer, localization of the critical set, and convergence of the steepest descent—are the subject of the companion paper [56].

On the mathematical side, the finite-dimensional theory of the quasi-norms that Section 2 revisits is part of the folklore of the quasi-Banach spaces [29], and the statements that hold on an arbitrary vector space are classical [32]; what we add is the comparison with the distance to the null space, which characterizes the quasi-seminorms within the CH class, and the structure theorem that follows from it. The star-shaped bodies and their radial functions are standard objects of convex and integral geometry [20 Chapter 0]. The conic splines are the piecewise-linear functions on a fan of toric geometry [19 Section 3.4] and, for d=3d = 3, the degree-one spherical splines of Alfeld, Neamtu and Schumaker [1, 2, 3], whose approximation order is due to Neamtu and Schumaker [42]. The context differs on two counts: spherical splines fit data that are given on the sphere S2S^2, a temperature map for instance, with a degree that is free, whereas we fit a quasi-norm, whose 1-homogeneity excludes the higher degrees but allows the theory to be formulated in an arbitrary dimension dd. What is specific to our setting is the identification of the non-negative conic splines with the polyhedral CH functionals, the reading of the nodal values as the parameters of a learning problem, the extension of the rate to the non-convex (star-shaped) bodies with a bound on the constant in terms of the Hessian of the gauge, and the second-order estimate under a measure-valued Hessian through the spherical Hessian variation, which is the restriction to the sphere of the Hessian–Schatten total variation of [9].

Organization of the Paper

The paper devotes one section to each of the three main issues, followed by the detailed treatment of the planar case. In Section 2, we identify the relevant class of amplitude-equivariant regularizers by linking it to the CH class (§ 3.1) and by providing a complete characterization of the most important members of the family: the quasi-norms and the quasi-seminorms, which offer control over the null space (§ 3.2). These functionals are the non-convex counterparts of the norms and the seminorms on Rd\mathbb{R}^d, respectively. We also make the correspondence with the star-shaped bodies and their gauges explicit (§ 3.3), which is the geometric language of the approximation results. In § 4, we turn to the parametrization: we recall the polyhedral fans and construct the conic B-splines, with their dual basis and their stability bounds (§ 4.1), and we prove the representation theorem for the polyhedral CH functionals (§ 4.2). In § 5, we establish the rates: after defining the regularity indices of a CH functional, including the spherical Hessian variation by duality (§ 5.1), we prove the uniform second-order estimate together with its sharpness (§ 5.2) and the L1L_1 estimate under a bounded spherical Hessian variation (§ 5.3). In § 6, we work out the details for the planar case d=2d=2, where everything becomes explicit: the conic B-splines on a uniform polar grid are trigonometric hat functions, the spherical Hessian variation of a polyhedral profile is an ℓ1\ell_1-norm of second differences of the nodal values, and the approximation constants are sharp.

Notations

The symbol X\mathcal{X} denotes an abstract vector space with generic elements x,y∈Xx,y\in\mathcal{X}. The Minkowski sum of two subsets B1,B2⊆XB_1,B_2\subseteq\mathcal{X} is B1+B2={x+y:x∈B1,y∈B2}B_1+B_2=\{x+y:x\in B_1,y\in B_2\}, and the rescaling of B⊆XB\subseteq\mathcal{X} by a scalar λ>0\lambda>0 is λB={λx:x∈B}\lambda B=\{\lambda x:x\in B\}; in particular, if B∥⋅∥={x∈X:∥x∥≤1}B_{\lVert\cdot\rVert}=\{x\in\mathcal{X}:\lVert x\rVert\leq1\} is the unit ball of a norm ∥⋅∥\lVert\cdot\rVert on X\mathcal{X}, then λB∥⋅∥={x∈X:∥x∥≤λ}\lambda B_{\lVert\cdot\rVert}=\{x\in\mathcal{X}:\lVert x\rVert\leq\lambda\} is its λ\lambda-inflated version.

When the setting is concrete with X=Rd\mathcal{X}=\mathbb{R}^d, we use boldface variables (x,y∈Rd)(\mathbf{x},\mathbf{y}\in\mathbb{R}^d) to remind us that the objects are “ordinary” vectors and that the underlying results are restricted to finite dimensions. The Euclidean norm of Rd\mathbb{R}^d is ∥⋅∥2\lVert\cdot\rVert_2, its unit ball is Bℓ2={x∈Rd:∥x∥2≤1}B_{\ell_2}=\{\mathbf{x}\in\mathbb{R}^d:\lVert\mathbf{x}\rVert_2\leq1\}, and the corresponding hypersphere is Sd−1={ξ∈Rd:∥ξ∥2=1}\mathbb{S}^{d-1}=\{\boldsymbol{\xi}\in\mathbb{R}^d:\lVert\boldsymbol{\xi}\rVert_2=1\}, so that a point x∈Rd∖{0}\mathbf{x}\in\mathbb{R}^d\setminus\{0\} is written in hyperpolar coordinates as x=rξ\mathbf{x}=r\boldsymbol{\xi} with r=∥x∥2∈R+=[0,∞)r=\lVert\mathbf{x}\rVert_2\in\mathbb{R}_+=[0,\infty) and ξ=x/∥x∥2∈Sd−1\boldsymbol{\xi}=\mathbf{x}/\lVert\mathbf{x}\rVert_2\in\mathbb{S}^{d-1}.

The radial function of a star-shaped body BB is denoted by rB(ξ)r_B(\boldsymbol{\xi}) (written ρB\rho_B in [20]): it is the radius of BB in the direction ξ∈Sd−1\boldsymbol{\xi}\in\mathbb{S}^{d-1}. If BqB_q is the unit ball of a quasi-norm qq, then rBq(ξ)=1/q(ξ)r_{B_q}(\boldsymbol{\xi})=1/q(\boldsymbol{\xi}) is the reciprocal of the angular profile q(ξ)q(\boldsymbol{\xi}) of qq. For d=2d=2, the latter is denoted by ϱ(θ)=q(ξ(θ))\varrho(\theta)=q(\boldsymbol{\xi}(\theta)) with ξ(θ)=(cos⁡θ,sin⁡θ)∈S1\boldsymbol{\xi}(\theta)=(\cos\theta,\sin\theta)\in\mathbb{S}^1.

The symbol Hk\mathcal{H}^k denotes the kk-dimensional Hausdorff measure—the surface measure of Sd−1\mathbb{S}^{d-1} when $k=d-1—and is not to be confused with the Hessian H{q}\mathrm{H}\{q\}.

Amplitude Equivariance and Quasi-norms

Amplitude Equivariance and the CH Class

Following [57], we say that a functional g:Rd→[0,∞)g : \mathbb{R}^{d} \to[0,\infty) is amplitude-equivariant if there exists a function a:R\{0}→(0,∞)a : \mathbb{R}\backslash\{0\} \to(0,\infty) such that

References

  1. [1]P. Alfeld, M. Neamtu, and L. L. Schumaker, Bernstein–Bézier polynomials on spheres and sphere-like surfaces, Computer Aided Geometric Design, 13 (1996), pp. 333–349.DOI
  2. [2]——, Dimension and local bases of homogeneous spline spaces, SIAM Journal on Mathematical Analysis, 27 (1996), pp. 1482–1501.DOI
  3. [3]——, Fitting scattered data on sphere-like surfaces using spherical splines, Journal of Computational and Applied Mathematics, 73 (1996), pp. 5–43.DOI
  4. [4]L. Ambrosio, N. Fusco, and D. Pallara, Functions of Bounded Variation and Free Discontinuity Problems, Oxford University Press, Oxford, 2000.
  5. [5]B. Amos, L. Xu, and J. Z. Kolter, Input convex neural networks, in Proceedings of the 34th International Conference on Machine Learning, vol. 70 of PMLR, 2017, pp. 146–155.arxiv.org/abs/1609.07152
  6. [6]T. Aoki, Locally bounded linear topological spaces, Proceedings of the Imperial Academy of Tokyo, 18 (1942), pp. 588–594.DOI
  7. [7]S. Arridge, P. Maass, O. Öktem, and C.-B. Schönlieb, Solving inverse problems using data-driven models, Acta Numerica, 28 (2019), pp. 1–174.DOI
  8. [8]K. Atkinson and W. Han, Spherical Harmonics and Approximations on the Unit Sphere: An Introduction, vol. 2044 of Lecture Notes in Mathematics, Springer, Berlin, 2012.DOI
  9. [9]S. Aziznejad, J. Campos, and M. Unser, Measuring complexity of learning schemes using Hessian–Schatten total variation, SIAM Journal on Mathematics of Data Science, 5 (2023), pp. 422–445.arxiv.org/abs/2112.06209
  10. [10]M. Benning and M. Burger, Modern regularization methods for inverse problems, Acta Numerica, 27 (2018), pp. 1–111.arxiv.org/abs/1801.09922
  11. [11]A. Boulkhemaïr, A. Chakib, and A. Nachaoui, On the Lipschitz continuity of the Minkowski gauge of star-shaped sets, Journal of Convex Analysis, 30 (2023), pp. 1–20.
  12. [12]E. J. Candès, M. B. Wakin, and S. P. Boyd, Enhancing sparsity by reweighted ℓ₁ minimization, Journal of Fourier Analysis and Applications, 14 (2008), pp. 877–905.arxiv.org/abs/0711.1612
  13. [13]R. Chartrand, Exact reconstruction of sparse signals via nonconvex minimization, IEEE Signal Processing Letters, 14 (2007), pp. 707–710.DOI
  14. [14]Y. Chen and T. Pock, Trainable nonlinear reaction diffusion: A flexible framework for fast and effective image restoration, IEEE Transactions on Pattern Analysis and Machine Intelligence, 39 (2017), pp. 1256–1272.DOI
  15. [15]F. Demengel, Fonctions à hessien borné, Annales de l’Institut Fourier, 34 (1984), pp. 155–190.DOI
  16. [16]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.
  17. [17]S. Foucart and M.-J. Lai, Sparsest solutions of underdetermined linear systems via ℓq-minimization for 0 < q ≤ 1, Applied and Computational Harmonic Analysis, 26 (2009), pp. 395–407.
  18. [18]S. Foucart and H. Rauhut, A Mathematical Introduction to Compressive Sensing, Applied and Numerical Harmonic Analysis, Birkhäuser/Springer, New York, 2013.DOI
  19. [19]W. Fulton, Introduction to Toric Varieties, vol. 131 of Annals of Mathematics Studies, Princeton University Press, Princeton, NJ, 1993.
  20. [20]R. J. Gardner, Geometric Tomography, vol. 58 of Encyclopedia of Mathematics and its Applications, Cambridge University Press, Cambridge, 2nd ed., 2006.DOI
  21. [21]A. Goujon, J. Campos, and M. Unser, Stable parameterization of continuous and piecewise-linear functions, Applied and Computational Harmonic Analysis, 70 (2024), p. 101581. [CHECK] volume, pages and year reconstructed from memory.DOI
  22. [22]A. Goujon, S. Neumayer, P. Bohra, S. Ducotterd, and M. Unser, A neural-network-based convex regularizer for inverse problems, IEEE Transactions on Computational Imaging, 9 (2023), pp. 781–795.arxiv.org/abs/2211.12461
  23. [23]A. Goujon, S. Neumayer, and M. Unser, Learning weakly convex regularizers for convergent image-reconstruction algorithms, SIAM Journal on Imaging Sciences, 17 (2024), pp. 91–115.arxiv.org/abs/2308.10542
  24. [25]P. M. Gruber, Aspects of approximation of convex bodies, in Handbook of Convex Geometry, Volume A, P. M. Gruber and J. M. Wills, eds., North-Holland, Amsterdam, 1993, pp. 319–345.DOI
  25. [26]A. Habring and M. Holler, Neural-network-based regularization methods for inverse problems in imaging, GAMM-Mitteilungen, 47 (2024), p. e202470004.arxiv.org/abs/2312.14849
  26. [27]A. Hauptmann, S. Mukherjee, C.-B. Schönlieb, and F. Sherry, Convergent regularization in inverse problems and linear plug-and-play denoisers, Foundations of Computational Mathematics, 25 (2025), pp. 1087–1120.
  27. [28]S. Hurault, A. Leclaire, and N. Papadakis, Gradient step denoiser for convergent plug-and-play, in International Conference on Learning Representations (ICLR), 2022.arxiv.org/abs/2110.03220
  28. [29]N. J. Kalton, Quasi-Banach spaces, in Handbook of the Geometry of Banach Spaces, W. B. Johnson and J. Lindenstrauss, eds., vol. 2, North-Holland, Amsterdam, 2003, pp. 1099–1130. [CHECK] volume, pages and year reconstructed from memory.DOI
  29. [30]U. S. Kamilov, C. A. Bouman, G. T. Buzzard, and B. Wohlberg, Plug-and-play methods for integrating physical and learned models in computational imaging: Theory, algorithms, and applications, IEEE Signal Processing Magazine, 40 (2023), pp. 85–97.DOI
  30. [31]E. Kobler, A. Effland, K. Kunisch, and T. Pock, Total deep variation: A stable regularization method for inverse problems, IEEE Transactions on Pattern Analysis and Machine Intelligence, 44 (2022), pp. 9163–9180.DOI
  31. [32]G. Köthe, Topological Vector Spaces I, vol. 159 of Grundlehren der mathematischen Wissenschaften, Springer, Berlin, 1983.DOI
  32. [33]K. Kunisch and T. Pock, A bilevel optimization approach for parameter learning in variational models, SIAM Journal on Imaging Sciences, 6 (2013), pp. 938–983.DOI
  33. [34]A. Lanza, S. Morigi, I. W. Selesnick, and F. Sgallari, Sparsity-inducing nonconvex nonseparable regularization for convex image processing, SIAM Journal on Imaging Sciences, 12 (2019), pp. 1099–1134.DOI
  34. [35]O. Leong, E. O’Reilly, and Y. S. Soh, The star geometry of critic-based regularizer learning, in Advances in Neural Information Processing Systems, vol. 37, 2024.
  35. [36]O. Leong, E. O’Reilly, Y. S. Soh, and V. Chandrasekaran, Optimal regularization for a data source, Foundations of Computational Mathematics, 26 (2026), pp. 889–938.arxiv.org/abs/2212.13597
  36. [37]H. Li, J. Schwab, S. Antholzer, and M. Haltmeier, NETT: solving inverse problems with deep neural networks, Inverse Problems, 36 (2020), p. 065005.arxiv.org/abs/1803.00092
  37. [38]S. Lunz, O. Öktem, and C.-B. Schönlieb, Adversarial regularizers in inverse problems, in Advances in Neural Information Processing Systems, vol. 31, 2018, pp. 8507–8516.arxiv.org/abs/1805.11572
  38. [39]T. Lyche and R. Winther, A stable recurrence relation for trigonometric B-splines, Journal of Approximation Theory, 25 (1979), pp. 266–279.DOI
  39. [40]M. McCann and M. Unser, Biomedical image reconstruction: From the foundations to deep neural networks, Foundations and Trends in Signal Processing, 13 (2019), pp. 280–359.
  40. [41]S. Mukherjee, A. Hauptmann, O. Öktem, M. Pereyra, and C.-B. Schönlieb, Learned reconstruction methods with convergence guarantees: A survey of concepts and applications, IEEE Signal Processing Magazine, 40 (2023), pp. 164–182.DOI
  41. [42]M. Neamtu and L. L. Schumaker, On the approximation order of splines on spherical triangulations, Advances in Computational Mathematics, 21 (2004), pp. 3–20.DOI
  42. [43]M. Nikolova, Analysis of the recovery of edges in images and signals by minimizing nonconvex regularized least-squares, Multiscale Modeling & Simulation, 4 (2005), pp. 960–991.DOI
  43. [44]M. Nikolova, M. K. Ng, and C.-P. Tam, Fast nonconvex nonsmooth minimization methods for image restoration and reconstruction, IEEE Transactions on Image Processing, 19 (2010), pp. 3073–3088.DOI
  44. [45]S. Rolewicz, On a certain class of linear metric spaces, Bulletin de l’Académie Polonaise des Sciences, 5 (1957), pp. 471–473.
  45. [46]Y. Romano, M. Elad, and P. Milanfar, The little engine that could: Regularization by denoising (RED), SIAM Journal on Imaging Sciences, 10 (2017), pp. 1804–1844.arxiv.org/abs/1611.02862
  46. [47]S. Roth and M. J. Black, Fields of experts, International Journal of Computer Vision, 82 (2009), pp. 205–229.DOI
  47. [48]L. I. Rudin, S. Osher, and E. Fatemi, Nonlinear total variation based noise removal algorithms, Physica D: Nonlinear Phenomena, 60 (1992), pp. 259–268.DOI
  48. [49]E. Ryu, J. Liu, S. Wang, X. Chen, Z. Wang, and W. Yin, Plug-and-play methods provably converge with properly trained denoisers, in Proceedings of the 36th International Conference on Machine Learning, vol. 97 of PMLR, 2019, pp. 5546–5557.arxiv.org/abs/1905.05406
  49. [50]I. J. Schoenberg, On trigonometric spline interpolation, Journal of Mathematics and Mechanics, 13 (1964), pp. 795–825.
  50. [51]S. Scholtes, Introduction to Piecewise Differentiable Equations, SpringerBriefs in Optimization, Springer, New York, 2012.DOI
  51. [52]I. Selesnick, Sparse regularization via convex analysis, IEEE Transactions on Signal Processing, 65 (2017), pp. 4481–4494.DOI
  52. [53]Z. Shumaylov, J. Budd, S. Mukherjee, and C.-B. Schönlieb, Weakly convex regularisers for inverse problems: Convergence of critical points and primal-dual optimisation, in Proceedings of the 41st International Conference on Machine Learning, vol. 235 of PMLR, 2024, pp. 45286–45314.arxiv.org/abs/2402.01052
  53. [54]E. Soubies, L. Blanc-Féraud, and G. Aubert, A continuous exact ℓ₀ penalty (CEL0) for least squares regularized problem, SIAM Journal on Imaging Sciences, 8 (2015), pp. 1607–1639.DOI
  54. [55]E. M. Stein and G. Weiss, Introduction to Fourier Analysis on Euclidean Spaces, Princeton University Press, Princeton, NJ, 1971.
  55. [56]M. Unser, Learning non-convex amplitude-equivariant regularizers. Preprint, Part II of the present series, 2026.
  56. [57]M. Unser and S. Ducotterd, Universal architectures for the learning of polyhedral norms and convex regularizers, SIAM Journal on Imaging Sciences, (2026).
  57. [58]S. V. Venkatakrishnan, C. A. Bouman, and B. Wohlberg, Plug-and-play priors for model based reconstruction, in 2013 IEEE Global Conference on Signal and Information Processing, 2013, pp. 945–948.DOI
  58. [59]C.-H. Zhang, Nearly unbiased variable selection under minimax concave penalty, The Annals of Statistics, 38 (2010), pp. 894–942.DOI
  59. [60]G. M. Ziegler, Lectures on Polytopes, vol. 152 of Graduate Texts in Mathematics, Springer, New York, 1995.

Paper details

vlnc:W_na5hem6c
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 2

downloads · 0

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

  • Mathematics

Claude

View statement
Looking for related papers…
Contents