Pith. sign in

REVIEW 3 major objections 7 minor 29 references

A Lightweight U-like Network Utilizing Neural Memory Ordinary Differential Equations for Slimming the Decoder

T0 review · 3 major / 7 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read The paper proposes three plug-and-play decoders obtained by discretizing neural memory ordinary differential equations with explicit Euler, Heun, and linear multistep methods, and claims they replace the decoders of U-like networks to cut…

desk verdict Efficiency numbers are credible, but the performance-parity claim rests on uncontrolled external baselines until matched re-training confirms it. read the letter →

arxiv 2412.06262 v1 pith:MTWFOJWA submitted 2024-12-09 cs.CV cs.AIeess.IV

Quansong He , Xiaojun Yao , Jun Wu , Zhang Yi , Tao He This is my paper · ORCID
classification cs.CVcs.AIeess.IV
keywords neuralmemoryordinarydifferentialequationsmedicalimagesegmentationlightweightdecoderU-likenetworksdiscretizedODEsolversskinlesionparameterreductionplug-and-play
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 establish that the decoder half of a U-like segmentation network can be replaced by a small, plug-in module derived from a neural memory ordinary differential equation (nmODE) without hurting accuracy and with large efficiency gains. Three discretizations of the nmODE—explicit Euler, Heun, and a linear multistep method—give three decoder variants (EED, HD, LMD). Across the PH2, ISIC2017, and ISIC2018 skin-lesion datasets, embedding these decoders in UNet, Att-UNet, UNeXt, MALUNet, and EGE-UNet reduces parameters by roughly 20–50% and FLOPs by up to 74%, while reported mIoU and Dice scores stay level or improve. If the pattern holds, the result matters because it offers a general way to slim existing U-like networks without redesigning their encoders or training from scratch.

What carries the argument

The central object is the discretized nmODE decoder: a neural memory ordinary differential equation (an ODE whose state $y(t)$ has a zero initial value and receives features from skip connections as external input $x(t)$) used to define the upward path of a U-like network. The work it does is to replace the decoder's many learned convolutional layers with a small number of parameterized skip-connection transforms $g(x_l,\theta_l)$ and a parameter-free state update $f(\cdot)$; the three numerical solvers—explicit Euler, Heun, and linear multistep—turn the continuous equation into the concrete layer recurrences that make the module plug-and-play.

What would settle it

Retrain one of the original networks (for example EGE-UNet on ISIC2018) under the exact same 7:3 random split, augmentation, optimizer, and 300-epoch schedule used for the modified decoders, and compare mIoU and Dice; if the retrained baseline matches or beats the nmODE-decoder version, the claim that accuracy is maintained loses its support.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that the decoder of a U-like network can be reformulated as an nmODE initial-value problem and solved by standard numerical integrators with a parameterless upward path. The nmODE state equation is $\dot{y}(t) = -y(t) + f\bigl(y(t) + g(x(t), \theta_t)\bigr)$, where skip-connection features $x(t)$ enter as sequential external inputs, the upward-path activations are the state $y(t)$, and $y(0)$ is initialized at zero instead of at the data. Discretizing this equation by explicit Euler, Heun, or linear multistep methods yields the layer recurrences $y_{l-1} = (1-\delta)y_l + \delta f(y_l + g(x_l,\theta_l))$ (EED), a predictor-corrector form (HD), and $y_{l-1} = y_{l+1} - 2\delta y_l + 2\delta f(y_l + g(x_l,\theta_l))$ (LMD). Because the $f$ function is parameter-free and the upward path shares weights, only the skip-connection $g$ functions carry parameters; this is what produces the 20–50% parameter reduction and up to 74% FLOP reduction reported on five U-like networks over three datasets.

Load-bearing premise

The performance claim assumes the baseline scores quoted from other papers were produced under training conditions equivalent enough to the authors' 7:3 split, 300-epoch schedule, and data augmentation that the comparison is fair.

Editorial extensions

If this is right

  • Att-UNet with the Euler decoder (EED) on PH2 drops from 34.88M parameters and 66.63 GFLOPs to 18.91M and 17.34 GFLOPs while mIoU rises from 0.858 to 0.905.
  • UNet with the linear-multistep decoder (LMD) on PH2 cuts parameters from 13.40M to 9.45M and raises mIoU from 0.878 to 0.905.
  • EGE-UNet with LMD on ISIC2018 cuts parameters from 0.053M to 0.041M and raises mIoU from 0.809 to 0.813, reaching reported state-of-the-art.
  • MALUNet with EED on ISIC2018 cuts parameters from 0.177M to 0.098M with mIoU rising from 0.803 to 0.806.
  • The three decoder variants are drop-in replacements: they require no changes to the encoder and no re-devised lightweight backbone, supporting the paper's claim of adaptability across U-like architectures.

Reading between the lines

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

  • If the parameterless upward path really transfers across architectures, the same decoder modules should also slim transformer-based U-shaped segmentation networks (for example Swin-UNet) without retraining their encoders; the paper only tests CNN- and MLP-based U-nets, so this is an extrapolation.
  • The ablation shows that the initial state's channel count is the main cost lever, so an obvious extension is an adaptive or grouped channel schedule that grows capacity only where the segmentation task needs it.
  • Heun's method, despite the extra g-function, was not consistently the best performer; this suggests the discretization's formal accuracy order may matter less than how skip-connection features are gated into the state, a hypothesis one could test by swapping the g-function design while holding the solver fixed.
  • Because the baselines come from other papers, the efficiency numbers are the solid part of the evidence while the accuracy-parity claim awaits a same-pipeline head-to-head; the authors' own ablations on UNet already show that the decoder algorithm, not just channel reduction, drives the gains.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 7 minor

Summary. The paper proposes replacing the decoder of U-shaped segmentation networks with lightweight blocks obtained by discretizing neural memory ordinary differential equations (nmODEs). Three discretizations are used: explicit Euler (EED), Heun (HD), and linear multistep (LMD). The upward path of the decoder is intended to be parameter-free, with trainable g-functions on the skip connections performing channel and spatial alignment. Experiments on PH2, ISIC2017, and ISIC2018 embed these decoders into UNet, Att-UNet, MALUNet, EGE-UNet, and UNeXt, reporting roughly 20–50% parameter reductions and up to 74% FLOP reductions while claiming that segmentation accuracy is maintained or slightly improved. Additional ablations on UNet/PH2 study the effect of the number of channels in the initial state y(0).

Significance. If the performance-parity and efficiency claims hold, the proposed decoders would be a practically useful plug-and-play way to compress existing U-like networks while preserving accuracy. The derivations of the three discretized blocks are explicit, the parameter and FLOP arithmetic in Tables 1–3 is transparent, and the ablation study in Table 2 shows internal consistency in isolating the decoder's contribution. The work is, however, largely an application of the previously published nmODE framework (Yi 2023; He et al. 2023; Wang et al. 2024) rather than a new theoretical construction. The main value rests on the empirical efficiency claim, which is currently weakened by the inconsistent baselines described in the major comments.

major comments (3)
  1. [§4.3, Table 1] The central claim that segmentation performance is maintained is not supported by the comparisons as reported. Modified networks were trained by the authors on a random 7:3 split with AdamW, CosineAnnealingLR, 300 epochs, batch size 8, and augmentation (§4.2), while most baseline rows are copied from Cheng et al. (2023), Wu et al. (2022), and Ruan et al. (2023), as the Table 1 footnote states. Those baselines need not use the same split, resolution, training schedule, or metric protocol. The discrepancy is visible within the paper: Table 1 lists UNet with DSC 0.919, whereas the authors' own “origin” UNet in Table 2 has DSC 0.9342 under the same input resolution. Likewise, Att-UNet(EED) exceeds the cited Att-UNet by 4.7 mIoU points on PH2, a margin large enough to be attributable to training setup rather than the decoder. The authors should retrain each original network under their own pipeline and compare matched pairs, or explicitly report the original networks' results under identical splits and settings.
  2. [§4.5, Table 2] The ablation table reports “origin” UNet with mIoU 0.8780 and DSC 0.9342, while Table 1 cites UNet with mIoU 0.878 and DSC 0.919. The near match in mIoU but clear mismatch in DSC indicates that the external baseline and the ablation baseline are not the same model or run. Table 2 therefore cannot serve as a matched control for the efficiency comparisons in Table 1. This internal inconsistency should be resolved by using one consistent UNet baseline across all tables.
  3. [§3.2, Function Selection & Initial Value Problem] The manuscript states that the upward path is parameter-free and that the f-function contains only batch normalization, which is described as a parameter-free operation. In standard PyTorch, BatchNorm2d has learnable weight and bias by default (affine=True). If the implementation uses affine=False, this should be stated explicitly and reflected in the code and parameter counts; if not, the statement that the decoder has no parameters on the upward path is inaccurate and the parameter savings in Tables 1–3 need to be recomputed or clarified to show they are unaffected.
minor comments (7)
  1. [§3.2, Eq. (7)] Equation (7) refers to ẏl without defining how this discrete derivative is computed inside the Heun block. From the derivation it appears to be the explicit-Euler intermediate value, but this should be stated in the text and labeled in Fig. 4(b).
  2. [§3.2, Eq. (6) and Eq. (7)] The notation “xn+1.θn+1” in Eq. (6) uses a period instead of a comma, and the index convention shifts to xl−1 and θl−1 in Eq. (7). Please align the indices so the reader can map the derivation onto the network layers.
  3. [§4.2] The text says CosineAnnealingLR is used “setting the maximum iteration count to 50” while training spans 300 epochs; these numbers are inconsistent. Please specify whether T_max is 50 epochs, 50 iterations, or something else.
  4. [Abstract and §4.3] The promised 20–50% parameter reduction and up to 74% FLOP reduction are presented as general statements, but they depend on comparisons with external baselines. A sentence noting that these figures use the reported baseline sources, and that matched retraining is needed for strict comparison, would make the claim more precise.
  5. [Table 1] Standard deviations are reported for the authors' models only. For visual fairness, either report mean±std for all baselines or state clearly that baseline values are single numbers taken from the original publications.
  6. [§4.3] The observation that HD underperforms EED and LMD despite having more parameters is mentioned but not discussed. Since this is surprising for a higher-order method, a brief explanation would help practitioners choose among the three decoders.
  7. [Figure 4] The caption says “the orange yl and red yl−1 are used as the upward path inputs and outputs,” but the diagram does not clearly identify which arrows are orange or red. Please add explicit labels in the figure.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the decoder designs are derived from a stated ODE model and standard discretizations, and the efficiency/accuracy claims are empirically benchmarked rather than reduced to their inputs by construction.

full rationale

The paper's derivation chain is not circular. The nmODE equation (Eq. 1) is adopted from prior work [Yi, 2023], and the three decoders are explicit discretizations of that equation using standard Euler, Heun, and linear multistep methods; the resulting update rules (Eqs. 4, 7, 10) are mathematical consequences, not restatements of the target claim. The parameter and FLOP reductions are arithmetic consequences of the decoder replacement, reported as measured quantities, and the 'maintaining performance' claim is supported by comparisons to external baseline numbers on public PH2, ISIC2017, and ISIC2018 benchmarks. Any concern that those external baselines may have used different training splits or schedules is a benchmarking-validity issue, not a circularity issue, because the baseline numbers are not generated by fitting the proposed model. Self-citations to [Yi, 2023], [He et al., 2023], [Wang et al., 2024], and [Hu et al., 2023] provide background and lineage for the nmODE framework, but they are not used as a uniqueness theorem, a forbidden alternative, or a fitted input, and the present experiments are independently evaluated. No step in the paper reduces, by definition or by self-citation, to its own inputs.

Assumptions & free parameters 1 free parameters · 3 assumptions · 0 invented entities

The central claim relies on the standard ODE solver theorems (Lipschitz continuity), the architectural assumption that a parameterless upward path with g-function transforms can replace decoders, and the generalization from three skin datasets to all U-like networks. One hyperparameter, the initial y-channel count, is tuned by ablation. No new entities are invented.

free parameters (1)
  • Initial y(0) channel count = 3 (default; ablation tested 1, 3, 8)
    Selected via ablation in Table 3; changes parameter count and accuracy. The paper chooses 3 as the default trade-off.
assumptions (3)
  • standard math Lipschitz continuity and existence and uniqueness of the ODE solution
    Invoked in Section 3.1 to justify solving Eq. (1); standard for ODE initial value problems.
  • domain assumption A U-like network decoder can be replaced by a parameterless upward path with shared parameters across layers, with skip-connection features reshaped by g-functions
    Section 3.2 states the upward path is parameter-free and only g-functions transform skip connections; this is the architectural bet that representational capacity is sufficient.
  • domain assumption Performance on PH2, ISIC2017, and ISIC2018 is representative enough to support the claim of adapting to all U-like networks
    The conclusion generalizes from five networks and three skin lesion datasets to all U-like networks without theoretical or broader empirical support.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Lightweight U-like Network Utilizing Neural Memory Ordinary Differential Equations for Slimming the Decoder." pith.science (2026). https://pith.science/paper/MTWFOJWA

@misc{pith2026241206262,
  author       = {Pith},
  title        = {Pith review of: A Lightweight U-like Network Utilizing Neural Memory Ordinary Differential Equations for Slimming the Decoder},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MTWFOJWA}},
  note         = {Machine review of arXiv:2412.06262}
}
read the original abstract

In recent years, advanced U-like networks have demonstrated remarkable performance in medical image segmentation tasks. However, their drawbacks, including excessive parameters, high computational complexity, and slow inference speed, pose challenges for practical implementation in scenarios with limited computational resources. Existing lightweight U-like networks have alleviated some of these problems, but they often have pre-designed structures and consist of inseparable modules, limiting their application scenarios. In this paper, we propose three plug-and-play decoders by employing different discretization methods of the neural memory Ordinary Differential Equations (nmODEs). These decoders integrate features at various levels of abstraction by processing information from skip connections and performing numerical operations on upward path. Through experiments on the PH2, ISIC2017, and ISIC2018 datasets, we embed these decoders into different U-like networks, demonstrating their effectiveness in significantly reducing the number of parameters and FLOPs while maintaining performance. In summary, the proposed discretized nmODEs decoders are capable of reducing the number of parameters by about 20% ~ 50% and FLOPs by up to 74%, while possessing the potential to adapt to all U-like networks. Our code is available at https://github.com/nayutayuki/Lightweight-nmODE-Decoders-For-U-like-networks.

Figures

Figures reproduced from arXiv: 2412.06262 by the authors.

Figure 1
Figure 1. Visualization of the mIoU results on the PH2 and [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. The U-like networks (with the proposed discretized [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Comparison between general NODE and nmODE. [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: (a) explicit Euler’s method discretized nmODEs decoder [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 5
Figure 5. Figure 5: Comparison of segmentation results between original networks and networks utilizing nmODEs decoders. Fig (a), (b), (c) and (d) [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

29 extracted references · 25 canonical work pages

  1. [5]

    Transformers for image recognition at scale

    [Dosovitskiy et al., 2020] A Dosovitskiy, L Beyer, A Kolesnikov, D Weissenborn, X Zhai, and T Un- terthiner. Transformers for image recognition at scale. arXiv preprint arXiv:2010.11929,

    arXiv 2020

  2. [8]

    Deep residual learning for image recog- nition,

    [He et al., 2015] Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recog- nition,

    work page 2015

  3. [10]

    Neue methoden zur approxi- mativen integration der differentialgleichungen einer un- abh¨angigen ver ¨anderlichen

    [Heun, 1900] Karl Heun. Neue methoden zur approxi- mativen integration der differentialgleichungen einer un- abh¨angigen ver ¨anderlichen. Z. Math. Phys , 45:23–38,

    work page 1900

  4. [12]

    [Liu et al., 2021] Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, and Baining Guo

    PMID: 37743765. [Liu et al., 2021] Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, and Baining Guo. Swin transformer: Hierarchical vision transformer using shifted windows. In Proceedings of the IEEE/CVF in- ternational conference on computer vision, pages 10012– 10022,

    work page 2021

  5. [15]

    At- tention u-net: Learning where to look for the pancreas,

    [Oktay et al., 2018] Ozan Oktay, Jo Schlemper, Loic Le Fol- goc, Matthew Lee, Mattias Heinrich, Kazunari Misawa, Kensaku Mori, Steven McDonagh, Nils Y Hammerla, Bernhard Kainz, Ben Glocker, and Daniel Rueckert. At- tention u-net: Learning where to look for the pancreas,

    work page 2018

  6. [16]

    Attractors in memory

    [Poucet and Save, 2005] Bruno Poucet and Etienne Save. Attractors in memory. Science, 308(5723):799–800,

    work page 2005

  7. [18]

    Malunet: A multi-attention and light-weight unet for skin lesion segmentation

    [Ruan et al., 2022] Jiacheng Ruan, Suncheng Xiang, Mingye Xie, Ting Liu, and Yuzhuo Fu. Malunet: A multi-attention and light-weight unet for skin lesion segmentation. In 2022 IEEE International Conference on Bioinformatics and Biomedicine (BIBM) , pages 1150–1156,

    work page 2022

  8. [19]

    EGE-UNet: an Efficient Group Enhanced UNet for skin lesion segmentation

    [Ruan et al., 2023] Jiacheng Ruan, Mingye Xie, Jingsheng Gao, Ting Liu, and Yuzhuo Fu. Ege-unet: an efficient group enhanced unet for skin lesion segmentation. arXiv preprint arXiv:2307.08473,

    work page Pith review arXiv 2023

Show all 29 references
  1. [20]

    Mlp-mixer: An all-mlp architecture for vision

    [Tolstikhin et al., 2021] Ilya O Tolstikhin, Neil Houlsby, Alexander Kolesnikov, Lucas Beyer, Xiaohua Zhai, Thomas Unterthiner, Jessica Yung, Andreas Steiner, Daniel Keysers, Jakob Uszkoreit, et al. Mlp-mixer: An all-mlp architecture for vision. Advances in neural infor- matio...

    2021

  2. [22]

    nmode-unet: A novel network for semantic segmentation of medical images

    [Wang et al., 2024] Shubin Wang, Yuanyuan Chen, and Zhang Yi. nmode-unet: A novel network for semantic segmentation of medical images. Applied Sciences, 14(1),

    2024

  3. [24]

    Wills, Colin Lever, Francesca Cacucci, Neil Burgess, and John O’Keefe

    [Wills et al., 2005] Tom J. Wills, Colin Lever, Francesca Cacucci, Neil Burgess, and John O’Keefe. Attractor dy- namics in the hippocampal representation of the local en- vironment. Science, 308(5723):873–876,

    2005

  4. [25]

    Fat-net: Fea- ture adaptive transformers for automated skin lesion seg- mentation

    [Wu et al., 2022] Huisi Wu, Shihuai Chen, Guilian Chen, Wei Wang, Baiying Lei, and Zhenkun Wen. Fat-net: Fea- ture adaptive transformers for automated skin lesion seg- mentation. Medical Image Analysis, 76:102327,

    2022

  5. [26]

    nmode: neural memory ordinary differ- ential equation

    [Yi, 2023] Zhang Yi. nmode: neural memory ordinary differ- ential equation. Artificial Intelligence Review, pages 1–36,

    2023

  6. [27]

    Road extraction by deep residual u-net

    [Zhang et al., 2018] Zhengxin Zhang, Qingjie Liu, and Yun- hong Wang. Road extraction by deep residual u-net. IEEE Geoscience and Remote Sensing Letters , 15(5):749–753,

    2018

  7. [28]

    Transfuse: Fusing transformers and cnns for med- ical image segmentation

    [Zhang et al., 2021] Yundong Zhang, Huiye Liu, and Qiang Hu. Transfuse: Fusing transformers and cnns for med- ical image segmentation. In Medical Image Computing and Computer Assisted Intervention–MICCAI 2021: 24th International Conference, Strasbourg, France, September 27–Oct...

    2021

  8. [29]

    Unet++: A nested u-net architecture for medical image segmenta- tion, 2018

    [Zhou et al., 2018] Zongwei Zhou, Md Mahfuzur Rahman Siddiquee, Nima Tajbakhsh, and Jianming Liang. Unet++: A nested u-net architecture for medical image segmenta- tion, 2018

    2018

  9. [1845]

    [Gao et al., 2022] Yunhe Gao, Mu Zhou, Ding Liu, and Dim- itris N. Metaxas. A multi-scale transformer for medical image segmentation: Architectures, model efficiency, and benchmarks. ArXiv, abs/2203.00131,

    2022 arXiv

  10. [1883]

    Swin-unet: Unet-like pure transformer for medical image segmentation

    [Cao et al., 2022] Cao, Hu, Yueyue Wang, Joy Chen, Dong- sheng Jiang, Xiaopeng Zhang, Qi Tian, and Manning Wang. Swin-unet: Unet-like pure transformer for medical image segmentation. In European conference on computer vision, pages 205–218,

    2022

  11. [1900]

    Enhancing robustness of medical image segmentation model with neural memory ordinary differ- ential equation

    [Hu et al., 2023] Junjie Hu, Chengrong Yu, Zhang Yi, and Haixian Zhang. Enhancing robustness of medical image segmentation model with neural memory ordinary differ- ential equation. International Journal of Neural Systems , 33(12):2350060,

    2023

  12. [2005]

    U-net: Convolutional networks for biomedical image segmentation

    [Ronneberger et al., 2015] Olaf Ronneberger, Philipp Fis- cher, and Thomas Brox. U-net: Convolutional networks for biomedical image segmentation. In MICCAI, pages 234–241,

    2015

  13. [2015]

    Med- ical image segmentation using discretized nmode

    [He et al., 2023] Quansong He, Tao He, and Zhang Yi. Med- ical image segmentation using discretized nmode. In 2023 CSIS-IAC, pages 497–503,

    2023

  14. [2016]

    Decoupled weight decay regularization

    [Loshchilov and Hutter, 2017] Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization. arXiv preprint arXiv:1711.05101,

    2017 arXiv

  15. [2017]

    Separable self-attention for mobile vision trans- formers,

    [Mehta and Rastegari, 2022] Sachin Mehta and Mohammad Rastegari. Separable self-attention for mobile vision trans- formers,

    2022

  16. [2018]

    Segnetr: Rethinking the local-global interactions and skip connections in u-shaped networks,

    [Cheng et al., 2023] Junlong Cheng, Chengrui Gao, Fengjie Wang, and Min Zhu. Segnetr: Rethinking the local-global interactions and skip connections in u-shaped networks,

    2023

  17. [2020]

    Augmented neural odes,

    [Dupont et al., 2019] Emilien Dupont, Arnaud Doucet, and Yee Whye Teh. Augmented neural odes,

    2019

  18. [2021]

    Unext: Mlp-based rapid medical im- age segmentation network

    [Valanarasu and Patel, 2022] Jeya Maria Jose Valanarasu and Vishal M Patel. Unext: Mlp-based rapid medical im- age segmentation network. In International Conference on Medical Image Computing and Computer-Assisted In- tervention, pages 23–33,

    2022

  19. [2022]

    Neural ordinary dif- ferential equations

    [Chen et al., 2018] Ricky TQ Chen, Yulia Rubanova, Jesse Bettencourt, and David K Duvenaud. Neural ordinary dif- ferential equations. Advances in neural information pro- cessing systems, 31,

    2018

  20. [2023]

    nmpls-net: Segmenting pulmonary lobes us- ing nmode

    [Dong et al., 2023] Peizhi Dong, Hao Niu, Zhang Yi, and Xiuyuan Xu. nmpls-net: Segmenting pulmonary lobes us- ing nmode. Mathematics, 11(22),

    2023

  21. [2024]

    Kevin Zhou, and Shuguang Cui

    [Wei et al., 2021] Jun Wei, Yiwen Hu, Ruimao Zhang, Zhen Li, S. Kevin Zhou, and Shuguang Cui. Shallow attention network for polyp segmentation,

    2021

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.