Pith. sign in

REVIEW 3 major objections 4 minor 300 references

Domain shift in plug-and-play reconstruction reduces to a measurable 'proximal mismatch' that directly controls convergence.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-02 00:46 UTC pith:AU6UFDEF

load-bearing objection A useful, well-run empirical study of proximal-matching adaptation for PnP, wrapped in a correct but not novel theory that has a genuine unverified assumption in the GS experiments. the 3 major comments →

arxiv 2607.14894 v1 pith:AU6UFDEF submitted 2026-07-16 eess.IV cs.LGmath.OC

Domain Adaptation of Mismatched Proximal Denoiser for Plug-and-Play Image Reconstruction

Guixian Xu , Jinglai Li , Junqi Tang This is my paper
classification eess.IV cs.LGmath.OC MSC 68U1065K10
keywords plug-and-playproximal mismatchconvergence analysisdomain adaptationimage reconstructionlearned proximal networksgradient-step denoisers
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

This paper studies plug-and-play proximal gradient descent (PnP-PGD) when the deployed denoiser—trained on one image domain—is used on a different target domain. It defines 'proximal mismatch' as the squared difference between the deployed denoiser and the target-domain reference proximal map at every PnP query point. Its main theorem bounds the average squared gradient of the target objective by O(1/K) plus a constant times the average squared proximal mismatch. This means persistent domain shift leaves an irreducible stationarity error, and reducing the average mismatch is the right adaptation target. The paper then shows that fine-tuning a denoiser by proximal matching—rather than by denoising MSE—reduces that mismatch and improves reconstruction substantially in few-shot settings, sometimes by more than 10 dB.

Core claim

The central claim is that the effect of domain shift on PnP-PGD is captured, to first order, by the sequence of proximal mismatch values d_k = || D_hat(z_k) - D_*(z_k) ||^2, where D_* is the proximal map of the target regularizer R_*. Theorem 3.1 shows that under smoothness and regularity assumptions, the average squared stationarity gap of the target objective decays as O(1/K) plus C_1 times the average squared mismatch. If the mismatch is square-summable the iterates converge to stationarity; if it is persistent, the stationarity error persists. This turns an abstract prior mismatch into a measurable quantity along the trajectory and provides a principled adaptation objective.

What carries the argument

The key object is the proximal mismatch d_k = || D_hat(z_k) - D_*(z_k) ||^2 between the deployed denoiser and the target proximal map at each PnP query point. It converts the deployed update into an inexact proximal step for the target objective. The analysis then uses the target proximal subproblem H_k(u) = (1/2)||u - z_k||^2 + R_*(u); its smoothness (L_H) and a gradient error inequality (Assumption 3.3) let the mismatch control both the descent loss and the stationarity residual. For Learned Proximal Networks and Gradient-Step denoisers, the paper verifies these conditions via explicit smoothness constants (1/alpha and 1/(1-L), respectively).

Load-bearing premise

The load-bearing premise is that the target domain has a well-defined proximal reference map D_* = prox_{R_*} and that every PnP query point satisfies the smoothness and unique-minimizer conditions of Assumption 3.3; the paper verifies these for ideal LPN and GS denoisers but not for the actually adapted ones.

What would settle it

Compute the stationarity gap of PnP-PGD on a controlled inverse problem with a known target regularizer, using a deployed denoiser deliberately mismatched to it; if the empirical average squared stationarity error does not grow approximately linearly with the average squared proximal mismatch as predicted, Theorem 3.1 is wrong. Equivalently, construct a denoiser with zero proximal mismatch at all query points that still fails to drive the stationarity measure to zero, disproving the direct implication.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • If the bound holds, a denoiser that is well-matched to the target proximal map will make PnP-PGD converge to a stationary point of the target objective; any persistent mismatch creates a floor on achievable stationarity.
  • The bound identifies average squared proximal mismatch along the PnP trajectory as the quantity to monitor and reduce during adaptation, rather than standalone denoising MSE.
  • Proximal-matching adaptation consistently outperforms MSE-based adaptation in the few-shot regime across two denoiser families and two inverse problems, with the largest gains (over 10 dB in severe shifts) at one or five target images.
  • In the matched case (deployed equals target reference), the bound recovers the standard O(1/K) stationarity rate of proximal gradient descent.

Where Pith is reading between the lines

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

  • The paper does not explore how to estimate the target proximal map when no target-trained reference is available; a natural extension is self-supervised or trajectory-based proximal estimation that directly minimizes the mismatch term.
  • A natural extension is to sample adaptation inputs from the actual PnP trajectory rather than only independent noisy images, which would directly reduce the mismatch term appearing in the bound.
  • The proximal-matching idea may transfer beyond imaging to other plug-and-play inexact proximal algorithms (e.g., RED or ADMM variants); a testable prediction is that reconstruction quality should correlate monotonically with the average squared mismatch, with the correlation strength depending on the constants in the bound.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The paper studies plug-and-play proximal gradient descent (PnP-PGD) when the deployed denoiser is mismatched to the target domain. It defines proximal mismatch as the squared output difference between the deployed denoiser and a target reference proximal map D_* = prox_{R_*}, and shows (Theorem 3.1) that the average squared gradient of the target objective is bounded by O(1/K) plus a constant times the average squared mismatch, under smoothness and regularity assumptions on the target proximal subproblems. The bound motivates a proximal-matching (PM) adaptation objective, which is implemented for learned proximal networks (LPNs) and gradient-step (GS) denoisers. Experiments on Gaussian deblurring and 4x super-resolution under a BreCaHAD-to-CelebA domain shift report that PM adaptation improves PSNR over MSE adaptation, especially in the few-shot regime, and reduces a family-matched trajectory gap relative to a target-trained reference.

Significance. If the assumptions hold, Theorem 3.1 is a clean and useful result: it quantifies how persistent denoiser mismatch degrades stationarity in PnP-PGD and justifies proximal matching as an adaptation principle. The proof is self-contained, the mismatch measure d_k is defined rather than fitted, and there is no circular derivation. The two denoiser families (LPN and GS) are well chosen because they realize proximal structure in different ways. The experimental findings, if validated, are practically significant: the reported gains over MSE adaptation in the one-shot and five-shot regimes are substantial. The main weaknesses are that the GS half of the validation depends on an unverified global contractivity condition, and the experimental gap metric compares against a target-trained denoiser of the same family rather than the true D_*, so the empirical claims are conditional on a surrogate reference.

major comments (3)
  1. [Section 3.4.2 / Proposition B.2 / Tables 2 and 3] The GS proximal interpretation requires that ∇g_{σ,⋆} is globally L-Lipschitz with L<1; otherwise the induced regularizer may not exist, H_k may be nonconvex or nonsmooth, and L_H=1/(1-L) in Eq. (26) is invalid. The paper only reports bL_max, an empirical local Jacobian estimate evaluated on adaptation points. For a neural-network potential, global Lipschitz constants can exceed 1 even when pointwise Jacobian estimates are below 1. Since GS is one of the two families used to validate the central claim, this unverified global condition makes the theoretical support for the GS experiments conditional. The authors should either verify the global contractivity bound for their trained potentials, restrict the GS empirical claims to a version where the condition is checked, or clearly state that Theorem 3.1 applies to GS only when the global condition holds.
  2. [Section 5.1, gap definition; Section 6] The trajectory gap is computed against a target-trained denoiser of the same family, i.e., D_{θ⋆}, not against the true D_* = prox_{R_*}. Thus the experiments measure agreement with a surrogate reference, not with the proximal map of the target regularizer. The observation that lower gap correlates with higher PSNR is therefore an within-family consistency result, not a direct empirical confirmation of Theorem 3.1. The paper acknowledges this in the conclusion, but the abstract and Section 1 state more strongly that 'proximal matching adaptation reduces the proximal mismatch'; that claim should be qualified. A test with a synthetic or analytically known R_* would strengthen the validation, or the wording should be adjusted to 'mismatch relative to the family-matched target-trained reference.'
  3. [Assumption 3.3 / Section 3.4.2] Assumption 3.3 also requires that at every deployed iterate the gradient-dominance inequality (17) holds, and for GS denoisers that every deployed output lies in Im(D_{σ,⋆}). For LPNs this is automatic, but for GS it is not verified. The paper does not report any check of these conditions along actual PnP-PGD trajectories. Since the theorem's conclusion depends on them, the experimental validation should include at least trajectory-level diagnostics (e.g., whether \hat D_σ(z_k) stays in the image of the target map, or whether the inequality in (17) is satisfied). Without such checks, the applicability of Theorem 3.1 to the GS experiments is not established.
minor comments (4)
  1. [General] No code or detailed training configuration is provided. Releasing code and specifying hyperparameters (e.g., λ_con, L_max, γ annealing schedule, number of corruptions M) would substantially improve reproducibility.
  2. [Section 4.2/4.3] The proximal-matching objectives in Eqs. (37) and (41) depend on a bandwidth γ, but the annealing schedule is not described. Please specify how γ is set or annealed in the experiments.
  3. [Tables 2 and 3] The caption states that SSIM and gap are reported in units of 10^{-2}, but the figures use different scales. Clarify the units consistently. Also, the bL_max values are described as 'empirical local Jacobian estimates'; state explicitly how they are computed.
  4. [Throughout] There are typographical issues: 'asproximal mismatch' in the abstract, broken author name 'V o' in the references, and some missing spaces. Also, reference 'Vert et al., 2026' appears to be a future work; verify its status.

Circularity Check

0 steps flagged

No significant circularity: Theorem 3.1 is a standard inexact proximal-gradient bound with mismatch defined a priori; adaptation and evaluation rely on external results and independent target-trained references.

full rationale

The paper's central bound (Theorem 3.1) is not circular. d_k is defined in Eq. (11) as ||\hat D(z_k)-D_*(z_k)||^2 for an a priori fixed target reference D_*=prox_{R_*}; Lemma 3.1 bounds the proximal subproblem gap by (L_H/2)d_k^2 using L_H-smoothness of H_k, and Theorem 3.1 follows from telescoping the descent inequality and the residual control in Lemmas 3.2-3.3. No parameter is fitted to make the bound hold: C0 and C1 are explicit constants in Eq. (20). The LPN and GS realizations import proximal-representation theorems (Fang et al. 2024; Hurault et al. 2022b) as external, non-self citations; the GS instance is conditional on an assumption (L<1, outputs in Im(D_{σ,*})) whose empirical verification is incomplete, but that is a correctness/validity concern, not circularity. The adaptation objective is motivated by Proposition 4.1, whose population-target claim rests on the external consistency result of Fang et al.; the experimental gap metric is evaluated against a separately trained target-reference denoiser, acknowledged in the conclusion as a surrogate. Thus no step reduces to its own inputs by construction.

Axiom & Free-Parameter Ledger

5 free parameters · 5 axioms · 2 invented entities

The central theorem is conditional on a structured target proximal reference and Assumption 3.3, both of which are idealized. The experimental validation uses a target-trained denoiser as a surrogate for D_*, adding an unverified link between theory and practice. The main free parameters are the LPN/GS structure constants and the PM/contractivity hyperparameters, none of which are fully reported.

free parameters (5)
  • LPN strong-convexity parameter alpha = alpha in (0,1), not reported
    LPN parameterization Psi_theta = psi_theta + (alpha/2)||x||^2 sets L_H = 1/alpha and therefore inflates C1 in Theorem 3.1; the specific value used is not given.
  • PM bandwidth gamma and annealing schedule = not reported
    The proximal-matching losses L_PM and L_GS-PM depend on gamma; the population minimizer approaches the MAP/proximal denoiser only as gamma tends to 0, and the schedule used in experiments is not specified.
  • GS contractivity penalty lambda_con and L_max = not reported
    The final GS objective includes lambda_con * R_con with L_max < 1; these are hand-chosen and determine whether the contractivity condition required for the proximal interpretation is approached.
  • Adaptation corruption noise sigma_adapt = not reported
    The adaptation set is generated as y = x + sigma_adapt * delta in eq. (27); the mismatch between sigma_adapt and the noise encountered on PnP trajectories affects whether PM targets the right proximal behavior.
  • Adaptation epochs and numbers of corruptions M = LPN 200 epochs, GS 20 epochs; M not stated
    These experimental settings are needed to reproduce the few-shot comparison but are only partially specified.
axioms (5)
  • domain assumption The target domain is representable by a structured proximal reference D_* = prox_{R_*}, with R_* differentiable at the deployed iterates.
    Section 2.1 and Assumption 3.1/6: without this there is no target objective F_* against which to measure stationarity.
  • ad hoc to paper Assumption 3.3: every proximal subproblem H_k has a unique minimizer, is uniformly L_H-smooth, and satisfies the gradient-dominance inequality at every deployed iterate.
    Section 3.3: verified only for ideal LPN/GS references, not for the adapted models used in experiments; this is the strongest structural premise.
  • ad hoc to paper For GS denoisers, every visited deployed output lies in Im(D_{sigma,*}), so the induced regularizer phi_{sigma,*} is finite and differentiable there.
    Section 3.4.2: the authors state this as an additional assumption and do not verify it numerically on the PnP trajectory.
  • domain assumption The population PM minimizer converges to the MAP/proximal denoiser as gamma -> 0, under the consistency conditions of Fang et al. (2024).
    Proposition 4.1: this external consistency result is needed to connect the PM loss to proximal/MAP behavior.
  • standard math Standard convex-analysis facts: the conjugate of an alpha-strongly convex function is 1/alpha-smooth; strongly monotone gradient maps are invertible; the descent lemma for L-smooth functions.
    Used throughout Appendix B and the proofs of Lemmas 3.1-3.3 and Theorem 3.1.
invented entities (2)
  • Target-domain reference proximal map D_* = prox_{R_*} no independent evidence
    purpose: Defines the target objective F_* and the proximal-mismatch metric d_k; the entire analysis is conditional on its existence.
    D_* is not observed in practice; the experiments substitute a target-trained denoiser of the same family, which the authors acknowledge in Section 6.
  • Gaussian-smoothed posterior mode target D*_{PM,gamma} no independent evidence
    purpose: Population minimizer of the proximal-matching loss; links PM adaptation to MAP/proximal denoisers as gamma -> 0.
    Introduced in Proposition 4.1; depends on consistency conditions borrowed from Fang et al. and is not directly measured in the experiments.

pith-pipeline@v1.3.0-alltime-deepseek · 17868 in / 17072 out tokens · 155640 ms · 2026-08-02T00:46:02.496932+00:00 · methodology

0 comments
read the original abstract

Plug-and-play proximal gradient descent (PnP-PGD) enables flexible image reconstruction by using denoisers as implicit priors. In practice, these denoisers are often deployed outside their training domains. Existing analyses establish convergence under structural assumptions on the deployed denoiser, such as requiring it to be a proximal map or a contraction. However, they do not measure how domain mismatch affects convergence of PnP-PGD. We define this effect as \emph{proximal mismatch}: the discrepancy between a deployed denoiser $\widehat{\mathsf D}$ and a target-domain reference map $\mathsf D_\star=\operatorname{prox}_{R_\star}$ associated with the underlying regularizer $R_\star$. Under this mismatch, each denoising update becomes an inexact proximal step for the target objective. We further derive a stationarity bound that decays at a rate of $\mathcal{O}(1/K)$, with an additive term proportional to the average squared proximal mismatch. This result motivates adaptation via proximal matching rather than MSE-based adaptation alone. We study this approach with two established denoiser families: learned proximal networks and gradient-step denoisers. Experiments on Gaussian deblurring and super-resolution under substantial domain shift show that proximal matching adaptation improves reconstruction quality significantly over MSE-based adaptation, yielding the largest numerical gains in the few-shot regime.

Figures

Figures reproduced from arXiv: 2607.14894 by Guixian Xu, Jinglai Li, Junqi Tang.

Figure 1
Figure 1. Figure 1: Representative images from the source and target domains. BreCaHAD contains histopathology patches, whereas CelebA contains aligned facial images, producing a substantial source-target shift in both semantic content and visual statistics. 5. Numerical Experiments Our experiments address three questions. First, how much does deploying a source-trained denoiser degrade PnP-PGD reconstruction on the target do… view at source ↗
Figure 2
Figure 2. Figure 2: Gaussian deblurring performance under the BreCaHAD-to-CelebA domain shift. PSNR and relative trajectory mismatch are shown as functions of the number of target-domain adaptation images. Proximal-matching adaptation yields the largest PSNR gains in the low-data regime and reduces the mis￾match with the corresponding target-trained reference. Higher PSNR and lower mismatch are better. For LPNs, proximal matc… view at source ↗
Figure 3
Figure 3. Figure 3: Representative Gaussian deblurring results for nadapt = 50 under the BreCaHAD-to-CelebA domain shift. The source-trained denoisers produce visible domain-shift artifacts. LPN-PM and GS-AdaPM recover sharper facial contours and fewer local distortions than their MSE-adapted counterparts. The boxes mark the enlarged regions shown below each reconstruction. 16 [PITH_FULL_IMAGE:figures/full_fig_p016_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: ×4 super-resolution performance under the BreCaHAD-to-CelebA domain shift. PSNR and relative trajectory mismatch are shown as functions of the number of target-domain adaptation images. Proximal-matching adaptation yields the largest PSNR gains in the low-data regime and reduces the mis￾match with the corresponding target-trained reference. Higher PSNR and lower mismatch are better. For LPNs, proximal matc… view at source ↗
Figure 5
Figure 5. Figure 5: Representative ×4 super-resolution results for nadapt = 50 under the BreCaHAD-to-CelebA domain shift. The proximal-matching variants recover sharper facial contours and fewer local distortions than the corresponding MSE-adapted models. The boxes mark the enlarged regions shown below each reconstruction. 19 [PITH_FULL_IMAGE:figures/full_fig_p019_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: examines the effect of LPN adaptation along the PnP-PGD trajectory. At the same adaptation budget, PM-adapted denoisers generally achieve higher PSNR across iterations than their MSE-adapted counterparts. The difference is largest with one or five target images and decreases as the adaptation set grows. This trajectory-level trend agrees with the final reconstruction results in Tables 2 and 3. The right-ha… view at source ↗
Figure 7
Figure 7. Figure 7: Representative Gaussian deblurring results for LPN-PM at different target-domain adaptation budgets. The largest visual improvement occurs between one and five adaptation images. Additional target images further reduce artifacts and refine facial contours and local details. 29 [PITH_FULL_IMAGE:figures/full_fig_p029_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Representative Gaussian deblurring results for GS-AdaPM at different target-domain adaptation budgets. The reconstructions change less between one and five images and improve more clearly when 25 target images are used. Larger adaptation sets recover sharper facial contours and finer local structure [PITH_FULL_IMAGE:figures/full_fig_p030_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Representative ×4 super-resolution results for LPN-PM at different target-domain adaptation budgets. The largest visual improvement occurs between one and five target images. Further adaptation produces smaller refinements in facial boundaries and local detail. 30 [PITH_FULL_IMAGE:figures/full_fig_p030_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Representative ×4 super-resolution results for GS-AdaPM at different target-domain adaptation budgets. The main visual improvement occurs when the adaptation budget increases from five to 25 images. Larger budgets produce more coherent facial boundaries and fewer local distortions. 31 [PITH_FULL_IMAGE:figures/full_fig_p031_10.png] view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

300 extracted references · 1 canonical work pages

  1. [1]

    Journal of digital imaging , volume=

    The Cancer Imaging Archive (TCIA): maintaining and operating a public information repository , author=. Journal of digital imaging , volume=. 2013 , publisher=

    2013

  2. [2]

    Advances in Neural Information Processing Systems , volume=

    Unsupervised learning from incomplete measurements for inverse problems , author=. Advances in Neural Information Processing Systems , volume=

  3. [3]

    IEEE Signal Processing Magazine , volume=

    Algorithm unrolling: Interpretable, efficient deep learning for signal and image processing , author=. IEEE Signal Processing Magazine , volume=. 2021 , publisher=

    2021

  4. [4]

    Inverse Problems , volume=

    Deep unrolling networks with recurrent momentum acceleration for nonlinear inverse problems , author=. Inverse Problems , volume=. 2024 , publisher=

    2024

  5. [5]

    hal-04820468 , year=

    Practical Operator Sketching Framework for Accelerating Iterative Data-Driven Solutions in Inverse Problems , author=. hal-04820468 , year=

  6. [6]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Equivariant plug-and-play image reconstruction , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  7. [7]

    SIAM Journal on Imaging Sciences , volume=

    A stochastic proximal alternating minimization for nonsmooth and nonconvex optimization , author=. SIAM Journal on Imaging Sciences , volume=. 2021 , publisher=

    2021

  8. [8]

    Inverse Problems , year=

    A guide to stochastic optimisation for large-scale inverse problems , author=. Inverse Problems , year=

  9. [9]

    Advances in Neural Information Processing Systems , pages=

    On lazy training in differentiable programming , author=. Advances in Neural Information Processing Systems , pages=

  10. [10]

    Advances in Neural Information Processing Systems , volume=

    Normalization layers are all that sharpness-aware minimization needs , author=. Advances in Neural Information Processing Systems , volume=

  11. [11]

    International Conference on Artificial Intelligence and Statistics , pages=

    Convex geometry of two-layer relu networks: Implicit autoencoding and interpretable models , author=. International Conference on Artificial Intelligence and Statistics , pages=. 2020 , organization=

    2020

  12. [12]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Robust equivariant imaging: a fully unsupervised framework for learning to image from noisy and partial measurements , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  13. [13]

    Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision , pages=

    Image restoration by deep projected GSURE , author=. Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision , pages=

  14. [14]

    IEEE Transactions on Computational Imaging , volume=

    An educated warm start for deep image prior-based micro CT reconstruction , author=. IEEE Transactions on Computational Imaging , volume=. 2022 , publisher=

    2022

  15. [15]

    ICASSP 2019-2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages=

    Image restoration using total variation regularized deep image prior , author=. ICASSP 2019-2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages=. 2019 , organization=

    2019

  16. [16]

    Proceedings of the IEEE/CVF International Conference on Computer Vision Workshops , pages=

    DeepRED: Deep image prior powered by RED , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision Workshops , pages=

  17. [17]

    IEEE Signal Processing Magazine , volume=

    Deep internal learning: Deep learning from a single input , author=. IEEE Signal Processing Magazine , volume=. 2024 , publisher=

    2024

  18. [18]

    Radon Series on Computational and Applied Mathematics , year=

    Unsupervised approaches based on optimal transport and convex analysis for inverse problems in imaging , author=. Radon Series on Computational and Applied Mathematics , year=

  19. [19]

    Mathematical Programming , volume=

    Parallel coordinate descent methods for big data optimization , author=. Mathematical Programming , volume=. 2016 , publisher=

    2016

  20. [20]

    Medical physics , volume=

    A three-dimensional statistical approach to improved image quality for multislice helical CT , author=. Medical physics , volume=. 2007 , publisher=

    2007

  21. [21]

    Journal of optimization theory and applications , volume=

    A primal--dual splitting method for convex optimization involving Lipschitzian, proximable and linear composite terms , author=. Journal of optimization theory and applications , volume=. 2013 , publisher=

    2013

  22. [22]

    Advances in Computational Mathematics , volume=

    A splitting algorithm for dual monotone inclusions involving cocoercive operators , author=. Advances in Computational Mathematics , volume=. 2013 , publisher=

    2013

  23. [23]

    arXiv preprint arXiv:2006.11630 , year=

    A Fast Stochastic Plug-and-Play ADMM for Imaging Inverse Problems , author=. arXiv preprint arXiv:2006.11630 , year=

    Pith/arXiv arXiv 2006

  24. [24]

    Set-valued and variational analysis , volume=

    A three-operator splitting scheme and its optimization applications , author=. Set-valued and variational analysis , volume=. 2017 , publisher=

    2017

  25. [25]

    Masters Thesis (Supervisor: Junqi Tang and Mike Davies), University of Edinburgh , year=

    Compressing Denoising Networks , author=. Masters Thesis (Supervisor: Junqi Tang and Mike Davies), University of Edinburgh , year=

  26. [26]

    arXiv preprint arXiv:2409.01985 , year=

    UNSURE: Unknown Noise level Stein's Unbiased Risk Estimator , author=. arXiv preprint arXiv:2409.01985 , year=

    Pith/arXiv arXiv

  27. [27]

    arXiv preprint arXiv:1903.08764 , year=

    Generalized Framework for Nonlinear Acceleration , author=. arXiv preprint arXiv:1903.08764 , year=

    Pith/arXiv arXiv 1903

  28. [28]

    SIAM Journal on Imaging Sciences (to appear) , year=

    Spring: A fast stochastic proximal alternating method for non-smooth non-convex optimization , author=. SIAM Journal on Imaging Sciences (to appear) , year=

  29. [29]

    arXiv preprint arXiv:2006.14798 , year=

    Training convolutional relu neural networks in polynomial time: Exact convex optimization formulations , author=. arXiv preprint arXiv:2006.14798 , year=

    Pith/arXiv arXiv 2006

  30. [30]

    IEEE/CVF Conference on Computer Vision and Pattern Recognition , year=

    The Neural Tangent Link Between CNN Denoisers and Non-Local Filters , author=. IEEE/CVF Conference on Computer Vision and Pattern Recognition , year=

  31. [31]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Equivariant imaging: Learning beyond the range space , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  32. [32]

    arXiv preprint arXiv:2312.11232 , year=

    Self-supervised learning for image super-resolution and deblurring , author=. arXiv preprint arXiv:2312.11232 , year=

    arXiv

  33. [33]

    arXiv preprint arXiv:2410.08646 , year=

    Fully Unsupervised Dynamic MRI Reconstruction via Diffeo-Temporal Equivariance , author=. arXiv preprint arXiv:2410.08646 , year=

    Pith/arXiv arXiv

  34. [34]

    arXiv preprint arXiv:2403.09327 , year=

    Perspective-Equivariant Imaging: an Unsupervised Framework for Multispectral Pansharpening , author=. arXiv preprint arXiv:2403.09327 , year=

    Pith/arXiv arXiv

  35. [35]

    Journal of Machine Learning Research , volume=

    Sensing theorems for unsupervised learning in linear inverse problems , author=. Journal of Machine Learning Research , volume=

  36. [36]

    International Conference on Machine Learning , pages=

    Gradient projection iterative sketch for large-scale constrained least-squares , author=. International Conference on Machine Learning , pages=. 2017 , organization=

    2017

  37. [37]

    SIAM Journal on Imaging Sciences , volume=

    NF-ULA: Normalizing Flow-Based Unadjusted Langevin Algorithm for Imaging Inverse Problems , author=. SIAM Journal on Imaging Sciences , volume=. 2024 , publisher=

    2024

  38. [38]

    SIAM Journal on Imaging Sciences , volume=

    Provably convergent plug-and-play quasi-Newton methods , author=. SIAM Journal on Imaging Sciences , volume=. 2024 , publisher=

    2024

  39. [39]

    IEEE Journal of Selected Topics in Signal Processing , volume=

    RARE: Image reconstruction using deep priors learned without groundtruth , author=. IEEE Journal of Selected Topics in Signal Processing , volume=. 2020 , publisher=

    2020

  40. [40]

    2023 , organization=

    3D PET-DIP reconstruction with relative difference prior using a SIRF-based objective , author=. 2023 , organization=

    2023

  41. [41]

    International Workshop on Machine Learning in Medical Imaging , pages=

    StoDIP: Efficient 3D MRF image reconstruction with deep image priors and stochastic iterations , author=. International Workshop on Machine Learning in Medical Imaging , pages=. 2024 , organization=

    2024

  42. [42]

    ICASSP 2024-2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages=

    Parameter-Efficient Adaptation for Computational Imaging , author=. ICASSP 2024-2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages=. 2024 , organization=

    2024

  43. [43]

    IEEE Transactions on Computational Imaging , volume=

    The practicality of stochastic optimization in imaging inverse problems , author=. IEEE Transactions on Computational Imaging , volume=. 2020 , publisher=

    2020

  44. [44]

    Proceedings of the 37th International Conference on Machine Learning , pages =

    Neural Networks are Convex Regularizers: Exact Polynomial-time Convex Optimization Formulations for Two-layer Networks , author =. Proceedings of the 37th International Conference on Machine Learning , pages =. 2020 , volume =

    2020

  45. [45]

    IEEE Transactions on Image Processing , volume=

    External Patch-Based Image Restoration Using Importance Sampling , author=. IEEE Transactions on Image Processing , volume=. 2019 , publisher=

    2019

  46. [46]

    arXiv preprint arXiv:2012.05169 , year=

    Convex Regularization Behind Neural Reconstruction , author=. arXiv preprint arXiv:2012.05169 , year=

    Pith/arXiv arXiv 2012

  47. [47]

    arXiv preprint arXiv:2006.02379 , year=

    CNN Denoisers As Non-Local Filters: The Neural Tangent Denoiser , author=. arXiv preprint arXiv:2006.02379 , year=

    Pith/arXiv arXiv 2006

  48. [48]

    arXiv preprint arXiv:2006.03224 , year=

    Scalable Plug-and-Play ADMM with Convergence Guarantees , author=. arXiv preprint arXiv:2006.03224 , year=

    Pith/arXiv arXiv 2006

  49. [49]

    Numerical Algorithms , volume=

    AIR Tools II: algebraic iterative reconstruction methods, improved implementation , author=. Numerical Algorithms , volume=. 2018 , publisher=

    2018

  50. [50]

    IEEE Transactions on Signal Processing , volume=

    A stochastic majorize-minimize subspace algorithm for online penalized least squares estimation , author=. IEEE Transactions on Signal Processing , volume=. 2017 , publisher=

    2017

  51. [51]

    Advances in Neural Information Processing Systems , pages=

    Nonlinear acceleration of stochastic algorithms , author=. Advances in Neural Information Processing Systems , pages=

  52. [52]

    arXiv preprint arXiv:2003.03910 , year=

    Geometry of First-Order Methods and Adaptive Acceleration , author=. arXiv preprint arXiv:2003.03910 , year=

    Pith/arXiv arXiv 2003

  53. [53]

    fast iterative shrinkage/thresholding algorithm

    On the convergence of the iterates of the “fast iterative shrinkage/thresholding algorithm” , author=. Journal of Optimization theory and Applications , volume=. 2015 , publisher=

    2015

  54. [54]

    Stochastic quasi-Fej

    Combettes, Patrick L and Pesquet, Jean-Christophe , journal=. Stochastic quasi-Fej. 2015 , publisher=

    2015

  55. [55]

    arXiv preprint arXiv:1903.01687 , year=

    Optimal algorithms for stochastic three-composite convex-concave saddle point problems , author=. arXiv preprint arXiv:1903.01687 , year=

    Pith/arXiv arXiv 1903

  56. [56]

    arXiv preprint arXiv:1901.07114 , year=

    Training neural networks as learning data-adaptive kernels: Provable representation and approximation benefits , author=. arXiv preprint arXiv:1901.07114 , year=

    Pith/arXiv arXiv 1901

  57. [57]

    arXiv preprint arXiv:1908.05355 , year=

    The generalization error of random features regression: Precise asymptotics and double descent curve , author=. arXiv preprint arXiv:1908.05355 , year=

    Pith/arXiv arXiv 1908

  58. [58]

    arXiv preprint arXiv:2003.01897 , year=

    Optimal Regularization Can Mitigate Double Descent , author=. arXiv preprint arXiv:2003.01897 , year=

    Pith/arXiv arXiv 2003

  59. [59]

    Proceedings of the National Academy of Sciences , volume=

    Reconciling modern machine-learning practice and the classical bias--variance trade-off , author=. Proceedings of the National Academy of Sciences , volume=. 2019 , publisher=

    2019

  60. [60]

    arXiv preprint arXiv:2003.02237 , year=

    Neural Kernels Without Tangents , author=. arXiv preprint arXiv:2003.02237 , year=

    Pith/arXiv arXiv 2003

  61. [61]

    Using the Nystr

    Williams, Christopher KI and Seeger, Matthias , booktitle=. Using the Nystr

  62. [62]

    Advances in neural information processing systems , pages=

    Random features for large-scale kernel machines , author=. Advances in neural information processing systems , pages=

  63. [63]

    Advances in neural information processing systems , pages=

    End-to-end kernel learning with supervised convolutional kernel networks , author=. Advances in neural information processing systems , pages=

  64. [64]

    Advances in Neural Information Processing Systems , pages=

    Generalization properties of learning with random features , author=. Advances in Neural Information Processing Systems , pages=

  65. [65]

    Yang, Tianbao and Li, Yu-Feng and Mahdavi, Mehrdad and Jin, Rong and Zhou, Zhi-Hua , booktitle=. Nystr

  66. [66]

    Advances in Neural Information Processing Systems , pages=

    Learning kernels with random features , author=. Advances in Neural Information Processing Systems , pages=

  67. [67]

    Artificial Intelligence and Statistics , pages=

    Deep kernel learning , author=. Artificial Intelligence and Statistics , pages=

  68. [68]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Learning adaptive random features , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  69. [69]

    Advances in Neural Information Processing Systems , pages=

    Stochastic variational deep kernel learning , author=. Advances in Neural Information Processing Systems , pages=

  70. [70]

    Advances in neural information processing systems , pages=

    On the global convergence of gradient descent for over-parameterized models using optimal transport , author=. Advances in neural information processing systems , pages=

  71. [71]

    arXiv preprint arXiv:1810.02054 , year=

    Gradient descent provably optimizes over-parameterized neural networks , author=. arXiv preprint arXiv:1810.02054 , year=

    Pith/arXiv arXiv

  72. [72]

    Advances in Neural Information Processing Systems , pages=

    Implicit regularization in matrix factorization , author=. Advances in Neural Information Processing Systems , pages=

  73. [73]

    Advances in Neural Information Processing Systems , pages=

    Graph neural tangent kernel: Fusing graph neural networks with graph kernels , author=. Advances in Neural Information Processing Systems , pages=

  74. [74]

    arXiv preprint arXiv:1911.00809 , year=

    Enhanced Convolutional Neural Tangent Kernels , author=. arXiv preprint arXiv:1911.00809 , year=

    Pith/arXiv arXiv 1911

  75. [75]

    Advances in neural information processing systems , pages=

    Learning and generalization in overparameterized neural networks, going beyond two layers , author=. Advances in neural information processing systems , pages=

  76. [76]

    Advances in Neural Information Processing Systems , pages=

    Learning overparameterized neural networks via stochastic gradient descent on structured data , author=. Advances in Neural Information Processing Systems , pages=

  77. [77]

    arXiv preprint arXiv:1811.03804 , year=

    Gradient descent finds global minima of deep neural networks , author=. arXiv preprint arXiv:1811.03804 , year=

    Pith/arXiv arXiv

  78. [78]

    arXiv preprint arXiv:1906.05827 , year=

    Kernel and deep regimes in overparametrized models , author=. arXiv preprint arXiv:1906.05827 , year=

    Pith/arXiv arXiv 1906

  79. [79]

    arXiv preprint arXiv:1902.01996 , year=

    Are all layers created equal? , author=. arXiv preprint arXiv:1902.01996 , year=

    Pith/arXiv arXiv 1902

  80. [80]

    arXiv preprint arXiv:2002.00585 , year=

    Proving the Lottery Ticket Hypothesis: Pruning is All You Need , author=. arXiv preprint arXiv:2002.00585 , year=

    Pith/arXiv arXiv 2002

Showing first 80 references.