Pith. sign in

REVIEW 3 major objections 6 minor 1 cited by

Comprehensive Examination of Unrolled Networks for Solving Linear Inverse Problems

T0 review · 3 major / 6 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read A memory-based unrolled network that reuses every past gradient and is trained with an unweighted sum of intermediate reconstruction errors achieves the best reconstruction quality among the tested unrolled designs.

desk verdict Useful broad ablation and a clean generalization (DeMUN), but no seed variance undercuts the 'uniformly' claims; accept for peer review, then revise. read the letter →

arxiv 2501.04608 v2 pith:RHHBIVRE submitted 2025-01-08 eess.IV cs.CVcs.LG

Eric Chen , Xi Chen , Arian Maleki , Shirin Jalali This is my paper · ORCID
classification eess.IVcs.CVcs.LG
keywords unrollednetworkslinearinverseproblemsimagereconstructiondeepmemorynetworkintermediatelossresidualconnectionsablationstudycompressedsensing
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

This paper tries to turn the design of unrolled networks for linear inverse problems from a costly search into a fixed recipe. It introduces DeMUN, an unrolled network whose update at every step is a learnable combination of the current estimate and the gradient formed at every previous step, so it contains standard projected gradient descent, momentum acceleration, and approximate message passing as special cases. Through extensive ablations on Gaussian and DCT measurement operators with natural-image data, the paper argues that training DeMUN with the unweighted intermediate loss, the sum of squared errors over all unrolled estimates, together with residual connections yields the best Peak Signal-to-Noise Ratio among all designs tested. It also reports practical rules of thumb: loss weighting that keeps at least 0.75 weight on early steps works as well as equal weighting, and projector depth between 5 and 15 layers hardly matters. If the recipe transfers to real imaging operators, practitioners can skip most design decisions.

What carries the argument

The machine that carries the argument is the Deep Memory Unrolled Network (DeMUN) update $$\tilde{x}_i = \alpha_i x_i + \sum_{j=0}^{i} \beta_i^j A^\top(y - A x_j),$$ followed by a learned projection $x_{i+1} = P_C(\tilde{x}_i)$ or the residual variant $x_{i+1} = \tilde{x}_i + P_C(\tilde{x}_i)$. The coefficients $\alpha_i, \beta_i^j$ are learned, so the data chooses how much of each past gradient to keep; the update reduces to standard gradient descent when only the current gradient survives and to momentum- or message-passing-style updates under other coefficient patterns. The second piece is the unweighted intermediate loss $\ell_{i,1} = \sum_{i=1}^{T} \|x_i - x^*\|_2^2$, which supervises every unrolled estimate instead of only the final one. Together they make the training landscape smoother and let extra unrolled steps keep improving the estimate, which underlies the headline performance claims.

What would settle it

Take a structured measurement operator from a real imaging modality, for example radial or Cartesian-sampled MRI with realistic receiver noise, and train the recommended configuration (DeMUN, unweighted intermediate loss, residual connections) against a standard unrolled baseline such as projected gradient descent with last-layer loss. If the recommended configuration fails to match or beat that baseline in reconstruction quality on held-out images, the paper's central recommendation would be shown not to transfer.

Watch

Extended reading notes

Core claim

The central discovery is that the choice of algorithm to unroll can be eliminated: DeMUN lets training learn the memory coefficients, and with an unweighted intermediate loss and residual connections it consistently outperforms unrolled projected gradient descent, momentum acceleration, and approximate message passing. With the last-layer loss, extra unrolled steps give almost no gain; with the intermediate loss, performance improves with more steps and plateaus rather than degrading. The paper also finds that weighting early layers slightly less (omega=0.95, 0.85) does not hurt, but weighting them much less (omega <= 0.5) or using skip-layer supervision degrades performance, and that increasing the number of projector layers from 5 to 15 is inconsequential while dropping to 3 hurts. These results hold for both Gaussian and DCT matrices, under additive noise, and across image resolutions from 32 by 32 to 80 by 80.

Load-bearing premise

The recipe was measured on Gaussian and DCT measurement operators acting on grayscale natural-image crops up to 80 by 80, and the paper's recommendations assume these findings transfer to other linear inverse problems, such as MRI or CT, where the forward operator is structured, the noise enters differently, and the image statistics differ.

Editorial extensions

If this is right

  • Training with the unweighted intermediate loss turns extra unrolled steps from a liability into a monotone, plateauing improvement, so a practitioner can pick a large step count without fear of overfitting.
  • The algorithm-choice decision disappears: because DeMUN contains projected gradient descent, momentum, and message-passing-style updates as special cases, the learned coefficients select the effective update, and the paper finds DeMUN matches or beats those baselines.
  • The loss-weighting choice is forgiving: any omega in [0.75, 1] performs about as well as the unweighted intermediate loss, so users do not need to tune the exponential weighting precisely.
  • Projector capacity is not a sensitive dial: 5 to 15 convolutional layers give nearly identical PSNR, including under noise and DCT operators, so small projectors can be used to save computation.
  • Robustness checks show the same design choices work for DCT matrices, additive Gaussian noise, and resolutions from 32 to 80, and reconstruction degrades more slowly than the input SNR.

Reading between the lines

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

  • The same recipe could be stress-tested on structured medical operators such as radial or Cartesian-sampled MRI and fan-beam CT; if it holds there, the paper's recipe becomes a default starting point for new imaging systems.
  • The slower-than-input-SNR degradation suggests the trained network behaves partly as a denoiser; a testable extension is to compare DeMUN's reconstruction error against state-evolution predictions of approximate message passing in the large-system limit.
  • The interaction between DeMUN's many parameters and the last-layer loss, where DeMUN sometimes loses to projected gradient descent, hints that intermediate supervision rescues high-capacity unrolled networks from poor local minima; one could test this by visualizing loss-landscape curvature or varying parameter count.
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

3 major / 6 minor

Summary. The paper studies design choices for unrolled networks solving linear inverse problems y = Ax + w. It introduces DeMUN, an unrolled architecture whose linear step is a learned combination of the current iterate and all past gradient terms A^T(y - Ax_j), implemented as trainable 1x1 convolutions; this generalizes PGD, Nesterov, and AMP in a single framework. The core empirical contribution is an extensive ablation across loss functions (last-layer, weighted intermediate, and skip-L intermediate losses), residual connections, the number of unrolled steps T ∈ {5, 15, 30}, DnCNN projector depth L ∈ {3, 5, 10, 15}, sampling rates {0.1, 0.2, 0.3, 0.4}n, Gaussian and DCT measurement matrices, additive Gaussian noise levels, and image resolutions from 32x32 to 80x80. Based on test PSNR on 2500 ImageNet validation crops, the authors propose four hypotheses and recommend DeMUN trained with the unweighted intermediate loss ℓ_{i,1} and residual connections.

Significance. If the empirical ordering is reproducible, the paper would give practitioners a useful shortcut: they could avoid sweeping over optimization algorithms, loss functions, and projector depths when designing unrolled networks. DeMUN is a clean and simple generalization, the ablation grid is broad, the DCT and noise robustness checks go beyond many ablation studies, and the code is publicly linked. The main limitations are statistical and interpretive: all comparisons come from single training runs, several gaps are tiny (0.01–0.2 dB), the word 'uniformly' is used without error bars, and Hypothesis 4 is internally contradicted by the paper's own DCT results. With repeated-training statistics and refined claims, this would be a valuable reference for the field.

major comments (3)
  1. [Sec. 4.3–4.4, Tables 4–6] The central claims that intermediate loss and residual connections 'uniformly' or 'consistently' improve performance are based on single training runs. Many reported differences are smaller than typical Adam run-to-run variation: for example, Table 5 (T=15, m=0.1n) shows PGD without residual at 26.42 dB and with residual at 26.43 dB, and Table 4 (m=0.1n) shows DeMUN at 26.09 vs 26.29 dB. No seeds, repeated runs, or confidence intervals are reported, so the pairwise ordering is not quantified. Because Hypotheses 1 and 2 use universal language, please report multi-seed means with standard deviations (or at least the number of seeds and observed range) and, where gaps are below roughly 0.1 dB, state that the difference is not statistically meaningful. This is load-bearing: reversal of even a few small-gap comparisons would weaken the 'uniformly' claims.
  2. [Sec. 4.6 and 5.3, Tables 10–12 and 18–20] Hypothesis 4 is not robust as stated. The hypothesis says that increasing L from 3 to 5 yields a 'noticeable improvement,' but Table 10 shows L=3 outperforming L=5 at m=0.1n (26.33 vs 26.29 dB) and Table 11 shows the same at m=0.4n (34.49 vs 34.43 dB). Moreover, in Section 5.3 the authors acknowledge that for DCT matrices 'in most cases... the performance gain from increasing L from 3 to 5 is also marginal,' which directly contradicts the unqualified claim in Hypothesis 4. Please either weaken the hypothesis (for example, 'L=3 can be slightly worse or comparable to L=5; L=5 is a safe default') or provide statistical evidence for a genuine improvement from L=3 to 5.
  3. [Sec. 4.3, Hypothesis 1 and Tables 1–3] The paper states that 'Among all algorithms that we have unrolled... DeMUN achieves the highest PSNR values when trained with the intermediate loss,' and the tables support that restricted claim. However, the broad wording of Hypothesis 1 and the abstract-level summary can be read as claiming DeMUN superiority regardless of loss. Under the last-layer loss, Tables 1–3 show PGD and Nesterov sometimes beating DeMUN (for example, Table 2, m=0.2n: PGD 27.71 vs DeMUN 27.23; Table 2, m=0.1n: PGD 24.72 vs DeMUN 24.42). The footnote acknowledges this, but the main text should make the restriction explicit in the hypothesis statement and in Section 6. Please rephrase to say that DeMUN superiority is demonstrated for the intermediate-loss configuration and not claimed in general.
minor comments (6)
  1. [Sec. 5.1] In the first bullet of Section 5.1, 'Tabel 13' should be 'Table 13'.
  2. [Sec. 2.1] The heading 'Loss F unction' contains an erroneous space; it should read 'Loss Function'.
  3. [Sec. 5.2] The additive noise is written as 'ω ∼ N(0, σ²I)', but ω already denotes the loss-weight parameter in Eq. (4); using a different symbol such as w would avoid confusion.
  4. [Sec. 5.4] The question 'Do we need more or fewer projections as we increase the number of projections?' should likely read 'as we increase the image resolution'.
  5. [Figures 4–13] The intermediate-PSNR figures would be easier to read with labeled axes (projection index vs PSNR) and a legend; currently some figures are visually similar and rely only on the caption.
  6. [Sec. 2.3] In the phrase 'theDeep Memory Unrolled Network' a space is missing after 'the'.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper's claims are empirical ablation findings tested on held-out configurations, with no fitted parameter or self-citation doubling as a prediction.

full rationale

The paper makes no claimed prediction or first-principles derivation that reduces to its own inputs. Its contributions are (i) a parametric architecture, DeMUN, whose update rule is defined in Eq. (6), with unit step-size PGD as the stated initialization (Appendix A.2.2), and (ii) four hypotheses that are explicitly described as summaries of simulations and then examined on additional measurement matrices, noise levels, and image resolutions. Trained network weights are fitted objects, but the paper does not rename a fitted parameter as a prediction: the reported PSNRs are direct evaluations on held-out test images, and the recommendations are empirical generalizations from those evaluations. Self-citations such as [5,6,8,9,42] concern background compressed-sensing and message-passing algorithms and are not load-bearing for the hypotheses. The paper's own limitations, including Section 4.6's admission that conclusions had not yet been validated beyond Gaussian matrices and the later DCT, noise, and resolution experiments, are scope limitations rather than circular steps. The lack of seed variance and error bars around single-run PSNR comparisons is a statistical robustness concern, not a circularity: it does not make any stated result equivalent to its inputs by construction. No equation in the paper is identical by definition to a claimed output, and no external result is imported through self-citation to force a design choice. Hence no circular step is identified.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

No universal constants are derived or fitted; the paper is an empirical study. The trained weights listed are model parameters, not additional free constants in a law. The domain assumptions are the standard linear forward model and the use of DnCNN as a projection proxy.

free parameters (2)
  • Memory coefficients α_i and β_i^j (DeMUN weights) = learned via ADAM during training
    These are the core trainable weights of the proposed DeMUN update (Eq. 6). They are fitted to the training data, but that is the method itself, not a hidden constant in an otherwise analytical derivation.
  • DnCNN projector weights = learned via ADAM
    Each projector is a DnCNN whose weights are trained end-to-end on the ImageNet-derived dataset; this is standard supervised learning rather than a hand-fitted parameter.
assumptions (3)
  • domain assumption The measurement process is well modeled by y = Ax + w with A known or known with small error.
    Stated in Section 1 as the starting model for all inverse problems considered.
  • domain assumption Natural images can be approximately projected onto a plausible image set by a DnCNN trained on ImageNet-derived grayscale crops.
    The paper replaces the unknown projection P_C with DnCNN (Section 4.1) and relies on this approximation for all experiments.
  • domain assumption ADAM training for 300 epochs with validation-based early stopping yields fair and comparable results across architectures.
    Training details in Appendix A.2.1; the comparison assumes this protocol does not bias one unrolled algorithm over another.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Comprehensive Examination of Unrolled Networks for Solving Linear Inverse Problems." pith.science (2026). https://pith.science/paper/RHHBIVRE

@misc{pith2026250104608,
  author       = {Pith},
  title        = {Pith review of: Comprehensive Examination of Unrolled Networks for Solving Linear Inverse Problems},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RHHBIVRE}},
  note         = {Machine review of arXiv:2501.04608}
}
read the original abstract

Unrolled networks have become prevalent in various computer vision and imaging tasks. Although they have demonstrated remarkable efficacy in solving specific computer vision and computational imaging tasks, their adaptation to other applications presents considerable challenges. This is primarily due to the multitude of design decisions that practitioners working on new applications must navigate, each potentially affecting the network's overall performance. These decisions include selecting the optimization algorithm, defining the loss function, and determining the number of convolutional layers, among others. Compounding the issue, evaluating each design choice requires time-consuming simulations to train, fine-tune the neural network, and optimize for its performance. As a result, the process of exploring multiple options and identifying the optimal configuration becomes time-consuming and computationally demanding. The main objectives of this paper are (1) to unify some ideas and methodologies used in unrolled networks to reduce the number of design choices a user has to make, and (2) to report a comprehensive ablation study to discuss the impact of each of the choices involved in designing unrolled networks and present practical recommendations based on our findings. We anticipate that this study will help scientists and engineers design unrolled networks for their applications and diagnose problems within their networks efficiently.

Figures

Figures reproduced from arXiv: 2501.04608 by the authors.

Figure 1
Figure 1. Diagram of projected gradient descent. Starting with [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. An example of the memory terms combined into a single image. [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. An example of the DnCNN architecture with [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (10 more)
Figure 4
Figure 4. Figure 4: DeMUN (no residual connections) with loss [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]
Figure 5
Figure 5. Figure 5: PGD (no residual connections) with loss ℓll. The networks are trained for T = 15 (left) and T = 30 (right), and the graph displays the PSNR after each intermediate projection [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]
Figure 6
Figure 6. Figure 6: DeMUN (no residual connections) with loss [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 7
Figure 7. Figure 7: PGD (no residual connections) with loss ℓi,1. The networks are trained for T = 15 (left) and T = 30 (right), and the graph displays the PSNR after each intermediate projection. From the above results and plots of the intermediate PSNR recovery, we make the following re…
Figure 8
Figure 8. Figure 8: DeMUN (including residual connections) with loss [PITH_FULL_IMAGE:figures/full_fig_p012_8.png]
Figure 9
Figure 9. Figure 9: PGD (including residual connections) with loss [PITH_FULL_IMAGE:figures/full_fig_p012_9.png]
Figure 10
Figure 10. Figure 10: DeMUN (including residual connections) with loss [PITH_FULL_IMAGE:figures/full_fig_p013_10.png]
Figure 11
Figure 11. Figure 11: DeMUN (including residual connections) with loss [PITH_FULL_IMAGE:figures/full_fig_p014_11.png]
Figure 12
Figure 12. Figure 12: DeMUN (including residual connections) with loss [PITH_FULL_IMAGE:figures/full_fig_p014_12.png]
Figure 13
Figure 13. Figure 13: DeMUN (including residual connections) with loss [PITH_FULL_IMAGE:figures/full_fig_p014_13.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

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

  1. How to warm-start your unfolding network

    cs.LG 2025-02 reject novelty 4.0 of 10

    Warm-starting an unfolding network via continuation and training with log-cosh is claimed to improve compressed sensing, but the supporting comparison is confounded by metric mismatch and test-set leakage.

Reference graph

Works this paper leans on

60 extracted references · 34 canonical work pages · cited by 1 Pith paper

  1. [1]

    D.L. Donoho. Compressed sensing. IEEE Transactions on Information Theory , 52(4):1289–1306, April 2006. ISSN 0018-9448. doi: 10.1109/TIT.2006.871582. URL http://ieeexplore.ieee. org/document/1614066/

    arXiv 2006

  2. [2]

    Michael Lustig, David Donoho, and John M. Pauly. Sparse MRI: The application of compressed sensing for rapid MR imaging. Magnetic Resonance in Medicine , 58(6):1182–1195, December

    work page

  3. [3]

    Candes and M.B

    E.J. Candes and M.B. Wakin. An Introduction To Compressive Sampling. IEEE Signal Processing Magazine, 25(2):21–30, March 2008. ISSN 1053-5888. doi: 10.1109/MSP.2007.914731. URL http://ieeexplore.ieee.org/document/4472240/

    arXiv 2008

  4. [4]

    Compressive sensing [lecture notes]

    Richard G Baraniuk. Compressive sensing [lecture notes]. IEEE signal processing magazine , 24 (4):118–121, 2007

    work page 2007

  5. [5]

    Donoho, Arian Maleki, and Andrea Montanari

    David L. Donoho, Arian Maleki, and Andrea Montanari. Message-passing algorithms for com- pressed sensing. Proceedings of the National Academy of Sciences, 106(45):18914–18919, November

    work page

  6. [6]

    Metzler, Arian Maleki, and Richard G

    Christopher A. Metzler, Arian Maleki, and Richard G. Baraniuk. From denoising to compressed sensing. IEEE Transactions on Information Theory , 62:5117–5144, 2016

    work page 2016

  7. [7]

    Venkatakrishnan, Charles A

    Singanallur V. Venkatakrishnan, Charles A. Bouman, and Brendt Wohlberg. Plug-and-Play priors for model based reconstruction. In 2013 IEEE Global Conference on Signal and Information Processing, pages 945–948, Austin, TX, USA, December 2013. IEEE. ISBN 978-1-4799-0248-4. doi: 10.1109/GlobalSIP.2013.6737048. URL http://ieeexplore.ieee.org/document/6737048/

    arXiv 2013

  8. [8]

    From compression to compressed sensing

    Shirin Jalali and Arian Maleki. From compression to compressed sensing. Applied and Computa- tional Harmonic Analysis , 40(2):352–385, 2016

    work page 2016

Show all 60 references
  1. [9]

    An efficient algorithm for compression-based compressed sensing

    Sajjad Beygi, Shirin Jalali, Arian Maleki, and Urbashi Mitra. An efficient algorithm for compression-based compressed sensing. Information and Inference: A Journal of the IMA , 8 (2):343–375, 2019

    2019

  2. [10]

    The Little Engine that Could: Regu- larization by Denoising (RED), September 2017

    Yaniv Romano, Michael Elad, and Peyman Milanfar. The Little Engine that Could: Regu- larization by Denoising (RED), September 2017. URL http://arxiv.org/abs/1611.02862. arXiv:1611.02862 [cs]

    2017 arXiv

  3. [11]

    Deep ADMM-Net for Compres- sive Sensing MRI

    Yan Yang, Jian Sun, Huibin Li, and Zongben Xu. Deep ADMM-Net for Compres- sive Sensing MRI. In D. Lee, M. Sugiyama, U. Luxburg, I. Guyon, and R. Gar- nett, editors, Advances in Neural Information Processing Systems , volume 29. Curran Asso- ciates, Inc., 2016. URL https://proc...

    2016

  4. [12]

    Baraniuk

    Ali Mousavi and Richard G. Baraniuk. Learning to invert: Signal recovery via Deep Convolutional Networks. In 2017 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 2272–2276, New Orleans, LA, March 2017. IEEE. ISBN 978-1-5090-4117-6. doi:...

    2017

  5. [13]

    Rick Chang, Chun-Liang Li, Barnabas Poczos, B.V.K

    J.H. Rick Chang, Chun-Liang Li, Barnabas Poczos, B.V.K. Vijaya Kumar, and Aswin C. Sankara- narayanan. One Network to Solve Them All — Solving Linear Inverse Problems Using Deep Projection Models. In 2017 IEEE International Conference on Computer Vision (ICCV) , pages 5889–589...

    2017

  6. [14]

    Baraniuk

    Ali Mousavi, Gautam Dasarathy, and Richard G. Baraniuk. DeepCodec: Adaptive Sensing and Recovery via Deep Convolutional Neural Networks, July 2017. URL http://arxiv.org/abs/ 1707.03386. arXiv:1707.03386 [stat]. 22

    2017 arXiv

  7. [15]

    Metzler, Ali Mousavi, and Richard G

    Christopher A. Metzler, Ali Mousavi, and Richard G. Baraniuk. Learned D-AMP: Principled Neural Network based Compressive Image Recovery, November 2017. URL http://arxiv.org/ abs/1704.06625. arXiv:1704.06625 [cs, stat]

    2017 arXiv

  8. [16]

    McCann, Kyong Hwan Jin, and Michael Unser

    Michael T. McCann, Kyong Hwan Jin, and Michael Unser. Convolutional Neural Networks for Inverse Problems in Imaging: A Review. IEEE Signal Processing Magazine, 34(6):85–95, Novem- ber 2017. ISSN 1053-5888. doi: 10.1109/MSP.2017.2739299. URL http://ieeexplore.ieee. org/document...

    2017

  9. [17]

    ISTA-Net: Interpretable Optimization-Inspired Deep Net- work for Image Compressive Sensing, June 2018

    Jian Zhang and Bernard Ghanem. ISTA-Net: Interpretable Optimization-Inspired Deep Net- work for Image Compressive Sensing, June 2018. URL http://arxiv.org/abs/1706.07929. arXiv:1706.07929 [cs]

    2018 arXiv

  10. [18]

    Unrolled Optimization with Deep Priors, December 2018

    Steven Diamond, Vincent Sitzmann, Felix Heide, and Gordon Wetzstein. Unrolled Optimization with Deep Priors, December 2018. URL http://arxiv.org/abs/1705.08041. arXiv:1705.08041 [cs]

    2018 arXiv

  11. [19]

    Hajnal, Anthony N

    Jo Schlemper, Jose Caballero, Joseph V. Hajnal, Anthony N. Price, and Daniel Rueckert. A Deep Cascade of Convolutional Neural Networks for Dynamic MR Image Reconstruction. IEEE Transactions on Medical Imaging , 37(2):491–503, February 2018. ISSN 0278-0062, 1558-254X. doi: 10.1...

    2018

  12. [20]

    Neumann Networks for Inverse Problems in Imaging, June 2019

    Davis Gilton, Greg Ongie, and Rebecca Willett. Neumann Networks for Inverse Problems in Imaging, June 2019. URL http://arxiv.org/abs/1901.03707. arXiv:1901.03707 [cs]

    2019 arXiv

  13. [21]

    Aggarwal, Merry P

    Hemant K. Aggarwal, Merry P. Mani, and Mathews Jacob. MoDL: Model-Based Deep Learn- ing Architecture for Inverse Problems. IEEE Transactions on Medical Imaging , 38(2):394– 405, February 2019. ISSN 0278-0062, 1558-254X. doi: 10.1109/TMI.2018.2865356. URL https://ieeexplore.iee...

    2019

  14. [22]

    Metzler, Richard G

    Gregory Ongie, Ajil Jalal, Christopher A. Metzler, Richard G. Baraniuk, Alexandros G. Dimakis, and Rebecca Willett. Deep Learning Techniques for Inverse Problems in Imaging. IEEE Journal on Selected Areas in Information Theory , 1(1):39–56, May 2020. ISSN 2641-8770. doi: 10.11...

    2020

  15. [23]

    Dave Van Veen, Ajil Jalal, Mahdi Soltanolkotabi, Eric Price, Sriram Vishwanath, and Alexan- dros G. Dimakis. Compressed Sensing with Deep Image Prior and Learned Regularization, Octo- ber 2020. URL http://arxiv.org/abs/1806.06438. arXiv:1806.06438 [stat]

    2020 arXiv

  16. [24]

    Deep Equilibrium Architectures for Inverse Problems in Imaging

    Davis Gilton, Gregory Ongie, and Rebecca Willett. Deep Equilibrium Architectures for Inverse Problems in Imaging. IEEE Transactions on Computational Imaging , 7:1123–1133, 2021. ISSN 2333-9403, 2334-0118, 2573-0436. doi: 10.1109/TCI.2021.3118944. URL https://ieeexplore. ieee.o...

    2021

  17. [25]

    Model Adaptation for Inverse Problems in Imaging

    Davis Gilton, Gregory Ongie, and Rebecca Willett. Model Adaptation for Inverse Problems in Imaging. IEEE Transactions on Computational Imaging , 7:661–674, 2021. ISSN 2333-9403, 2334-0118, 2573-0436. doi: 10.1109/TCI.2021.3094714. URL https://ieeexplore.ieee.org/ document/9477112/

    2021

  18. [26]

    Stochastic solutions for linear inverse problems using the prior implicit in a denoiser

    Zahra Kadkhodaie and Eero Simoncelli. Stochastic solutions for linear inverse problems using the prior implicit in a denoiser. In M. Ranzato, A. Beygelzimer, Y. Dauphin, P.S. Liang, and J. Wortman Vaughan, editors, Advances in Neural Information Processing Systems , volume 34,...

    2021

  19. [27]

    Shastri, Rizwan Ahmad, Christopher A

    Saurav K. Shastri, Rizwan Ahmad, Christopher A. Metzler, and Philip Schniter. Denoising Generalized Expectation-Consistent Approximation for MR Image Recovery. IEEE Journal on Selected Areas in Information Theory , 3(3):528–542, September 2022. ISSN 2641-8770. doi: 10.1109/JSA...

    2022

  20. [28]

    Beyond First-Order Tweedie: Solving Inverse Problems using Latent Diffusion, De- cember 2023

    Litu Rout, Yujia Chen, Abhishek Kumar, Constantine Caramanis, Sanjay Shakkottai, and Wen- Sheng Chu. Beyond First-Order Tweedie: Solving Inverse Problems using Latent Diffusion, De- cember 2023. URL http://arxiv.org/abs/2312.00852. arXiv:2312.00852 [cs]

    2023 arXiv

  21. [29]

    Physics-Inspired Compressive Sensing: Beyond deep unrolling

    Jian Zhang, Bin Chen, Ruiqin Xiong, and Yongbing Zhang. Physics-Inspired Compressive Sensing: Beyond deep unrolling. IEEE Signal Processing Magazine , 40(1):58–72, January 2023. ISSN 1053-5888, 1558-0792. doi: 10.1109/MSP.2022.3208394. URL https://ieeexplore.ieee.org/ document...

    2023

  22. [30]

    Plug-and- play methods for integrating physical and learned models in computational imaging: Theory, algorithms, and applications

    Ulugbek S Kamilov, Charles A Bouman, Gregery T Buzzard, and Brendt Wohlberg. Plug-and- play methods for integrating physical and learned models in computational imaging: Theory, algorithms, and applications. IEEE Signal Processing Magazine , 40(1):85–97, 2023

    2023

  23. [31]

    Block coordinate plug-and-play methods for blind inverse problems

    Weijie Gan, Yuyang Hu, Jiaming Liu, Hongyu An, Ulugbek Kamilov, et al. Block coordinate plug-and-play methods for blind inverse problems. Advances in Neural Information Processing Systems, 36, 2024

    2024

  24. [32]

    Ptychodv: Vision transformer-based deep unrolling network for ptychographic image reconstruction

    Weijie Gan, Qiuchen Zhai, Michael Thompson McCann, Cristina Garcia Cardona, Ulugbek S Kamilov, and Brendt Wohlberg. Ptychodv: Vision transformer-based deep unrolling network for ptychographic image reconstruction. IEEE Open Journal of Signal Processing , 2024

    2024

  25. [33]

    Yuyang Hu, Albert Peng, Weijie Gan, Peyman Milanfar, Mauricio Delbracio, and Ulugbek S. Kamilov. Stochastic Deep Restoration Priors for Imaging Inverse Problems, October 2024. URL http://arxiv.org/abs/2410.02057. arXiv:2410.02057 [eess]

    2024 arXiv

  26. [34]

    Decomposed Diffusion Sampler for Acceler- ating Large-Scale Inverse Problems, February 2024

    Hyungjin Chung, Suhyeon Lee, and Jong Chul Ye. Decomposed Diffusion Sampler for Acceler- ating Large-Scale Inverse Problems, February 2024. URL http://arxiv.org/abs/2303.05754. arXiv:2303.05754 [cs]

    2024 arXiv

  27. [35]

    Practical Compact Deep Compressed Sensing, November 2024

    Bin Chen and Jian Zhang. Practical Compact Deep Compressed Sensing, November 2024. URL http://arxiv.org/abs/2411.13081. arXiv:2411.13081 [cs]

    2024 arXiv

  28. [36]

    Self-Supervised Scalable Deep Compressed Sensing, August 2024

    Bin Chen, Xuanyu Zhang, Shuai Liu, Yongbing Zhang, and Jian Zhang. Self-Supervised Scalable Deep Compressed Sensing, August 2024. URL http://arxiv.org/abs/2308.13777. arXiv:2308.13777 [eess]

    2024 arXiv

  29. [37]

    MRI recovery with self- calibrated denoisers without fully-sampled data

    Muhammad Shafique, Sizhuo Liu, Philip Schniter, and Rizwan Ahmad. MRI recovery with self- calibrated denoisers without fully-sampled data. Magnetic Resonance Materials in Physics, Bi- ology and Medicine , October 2024. ISSN 1352-8661. doi: 10.1007/s10334-024-01207-1. URL https...

    2024 doi

  30. [38]

    Kamilov, and Brendt Wohlberg

    Weijie Gan, Qiuchen Zhai, Michael Thompson McCann, Cristina Garcia Cardona, Ulugbek S. Kamilov, and Brendt Wohlberg. PtychoDV: Vision Transformer-Based Deep Unrolling Network for Ptychographic Image Reconstruction, March 2024. URLhttp://arxiv.org/abs/2310.07504. arXiv:2310.075...

    2024

  31. [39]

    ReconNet: Non-Iterative Reconstruction of Images from Compressively Sensed Measurements

    Kuldeep Kulkarni, Suhas Lohit, Pavan Turaga, Ronan Kerviche, and Amit Ashok. ReconNet: Non-Iterative Reconstruction of Images from Compressively Sensed Measurements. In 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , pages 449–458, Las Vegas, NV, USA, ...

    2016

  32. [40]

    Learning fast approximations of sparse coding

    Karol Gregor and Yann LeCun. Learning fast approximations of sparse coding. In Proceedings of the 27th International Conference on International Conference on Machine Learning , ICML’10, page 399–406, Madison, WI, USA, 2010. Omnipress. ISBN 9781605589077

    2010

  33. [41]

    Vishal Monga, Yuelong Li, and Yonina C. Eldar. Algorithm Unrolling: Interpretable, Efficient Deep Learning for Signal and Image Processing, August 2020. URL http://arxiv.org/abs/ 1912.10557. arXiv:1912.10557 [eess]

    2020 arXiv

  34. [42]

    Compression-based compressed sensing

    Farideh E Rezagah, Shirin Jalali, Elza Erkip, and H Vincent Poor. Compression-based compressed sensing. IEEE Transactions on Information Theory , 63(10):6735–6752, 2017. 24

    2017

  35. [43]

    Neural Proximal Gradient Descent for Compressive Imaging, June

    Morteza Mardani, Qingyun Sun, Shreyas Vasawanala, Vardan Papyan, Hatef Monajemi, John Pauly, and David Donoho. Neural Proximal Gradient Descent for Compressive Imaging, June

  36. [44]

    Yuelong Li, Or Bar-Shira, Vishal Monga, and Yonina C. Eldar. Deep Algorithm Unrolling for Biomedical Imaging, August 2021. URL http://arxiv.org/abs/2108.06637. arXiv:2108.06637 [cs]

    2021 arXiv

  37. [45]

    A Fast Iterative Shrinkage-Thresholding Algorithm for Linear In- verse Problems

    Amir Beck and Marc Teboulle. A Fast Iterative Shrinkage-Thresholding Algorithm for Linear In- verse Problems. SIAM Journal on Imaging Sciences , 2(1):183–202, 2009. doi: 10.1137/080716542. URL https://doi.org/10.1137/080716542. eprint: https://doi.org/10.1137/080716542

    2009 doi

  38. [46]

    GSISTA-Net: generalized structure ISTA networks for image compressed sensing based on optimized unrolling algorithm

    Chunyan Zeng, Yan Yu, Zhifeng Wang, Shiyan Xia, Hao Cui, and Xiangkui Wan. GSISTA-Net: generalized structure ISTA networks for image compressed sensing based on optimized unrolling algorithm. Multimedia Tools and Applications , March 2024. ISSN 1573-7721. doi: 10.1007/ s11042-...

    2024 doi

  39. [47]

    AMP-Net: Denoising-Based Deep Unfolding for Compressive Image Sensing

    Zhonghao Zhang, Yipeng Liu, Jiani Liu, Fei Wen, and Ce Zhu. AMP-Net: Denoising-Based Deep Unfolding for Compressive Image Sensing. IEEE Transactions on Image Processing , 30: 1487–1500, 2021. ISSN 1057-7149, 1941-0042. doi: 10.1109/TIP.2020.3044472. URL https: //ieeexplore.iee...

    2021

  40. [48]

    Convolutional Neural Networks With Intermediate Loss for 3D Super-Resolution of CT and MRI Scans

    Mariana-Iuliana Georgescu, Radu Tudor Ionescu, and Nicolae Verga. Convolutional Neural Networks With Intermediate Loss for 3D Super-Resolution of CT and MRI Scans. IEEE Access, 8:49112–49124, 2020. ISSN 2169-3536. doi: 10.1109/ACCESS.2020.2980266. URL https://ieeexplore.ieee.o...

    2020

  41. [49]

    Deep Generalized Unfolding Networks for Image Restoration

    Chong Mou, Qian Wang, and Jian Zhang. Deep Generalized Unfolding Networks for Image Restoration. In 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 17378–17389, New Orleans, LA, USA, June 2022. IEEE. ISBN 978-1-66546- 946-3. doi: 10.1109/CVPR5...

    2022

  42. [50]

    Rethinking the inception architecture for computer vision

    Christian Szegedy, Vincent Vanhoucke, Sergey Ioffe, Jon Shlens, and Zbigniew Wojna. Rethinking the inception architecture for computer vision. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 2818–2826, 2016

    2016

  43. [51]

    Beyond a Gaussian Denoiser: Residual Learning of Deep CNN for Image Denoising

    Kai Zhang, Wangmeng Zuo, Yunjin Chen, Deyu Meng, and Lei Zhang. Beyond a Gaussian Denoiser: Residual Learning of Deep CNN for Image Denoising. IEEE Transactions on Image Processing, 26(7):3142–3155, July 2017. ISSN 1057-7149, 1941-0042. doi: 10.1109/TIP.2017. 2662206. URL http...

    2017

  44. [52]

    Deep Residual Learning for Image Recognition, December 2015

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep Residual Learning for Image Recognition, December 2015. URL http://arxiv.org/abs/1512.03385. arXiv:1512.03385 [cs]

    2015 arXiv

  45. [53]

    Identity Mappings in Deep Residual Networks, July 2016

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Identity Mappings in Deep Residual Networks, July 2016. URL http://arxiv.org/abs/1603.05027. arXiv:1603.05027 [cs]

    2016 arXiv

  46. [54]

    Convolu- tional neural networks: an overview and application in radiology

    Rikiya Yamashita, Mizuho Nishio, Richard Kinh Gian Do, and Kaori Togashi. Convolu- tional neural networks: an overview and application in radiology. Insights into Imaging , 9 (4):611–629, August 2018. ISSN 1869-4101. doi: 10.1007/s13244-018-0639-9. URL https: //insightsimaging...

    2018 doi

  47. [58]

    Sub- sequently, we normalize A according to A′ = A/∥A∥∞,2 to obtain our sampling matrix

    We calculate the maximum ℓ2 row norm of the matrix: ∥A∥∞,2 = max1≤i≤m qPn j=1 a2 ij. Sub- sequently, we normalize A according to A′ = A/∥A∥∞,2 to obtain our sampling matrix

  48. [59]

    For each loss reducer i, we initialize the first two terms αi, βi i to 1 and set all other weights to 0

  49. [60]

    All other network projector weights {P t C}t are initialized randomly [12, 17, 47]. For instance in PyTorch, convolutional weights are initialized by sampling from Uniform( − √ k, − √ k), where k = groups Cin · Q1 i=0 kernelsize i Here, groups specifies independent channel gro...

    2012

  50. [2007]

    doi: 10.1002/mrm.21391

    ISSN 0740-3194, 1522-2594. doi: 10.1002/mrm.21391. URL https://onlinelibrary. wiley.com/doi/10.1002/mrm.21391

    doi

  51. [2009]

    doi: 10.1073/pnas.0909892106

    ISSN 0027-8424, 1091-6490. doi: 10.1073/pnas.0909892106. URL https://pnas.org/doi/ full/10.1073/pnas.0909892106

    doi

  52. [2018]

    arXiv:1806.03963 [cs]

    URL http://arxiv.org/abs/1806.03963. arXiv:1806.03963 [cs]

    arXiv

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.