pith. sign in

arxiv: 2405.03420 · v2 · submitted 2024-05-06 · 💻 cs.CV · cs.AI

Implantable Adaptive Cells: A Novel Enhancement for Pre-Trained U-Nets in Medical Image Segmentation

Emil Benedykciuk , Marcin Denkowski , Grzegorz W\'ojcik This is my paper

Pith reviewed 2026-05-24 01:40 UTC · model grok-4.3

classification 💻 cs.CV cs.AI
keywords medical image segmentationU-Netneural architecture searchskip connectionsimplantable adaptive cellpre-trained modelsDARTS
0
0 comments X

The pith

Injecting Implantable Adaptive Cells into skip connections of pre-trained U-Nets improves segmentation accuracy by about 5 percentage points without full retraining.

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

This paper introduces Implantable Adaptive Cells, small modules found using Partially-Connected DARTS, that can be injected into the skip connections of already trained U-Nets for medical image segmentation. The approach aims to refine existing models without the need for complete retraining. On four datasets involving MRI and CT images, the method delivers consistent gains of roughly 5 percentage points in accuracy, with peaks at 11 points. It positions itself as a practical way to upgrade performance at lower cost than building new models from scratch.

Core claim

The paper establishes that small adaptive modules discovered by a gradient-based NAS technique can be implanted into the skip connections of pre-trained U-shaped models to enhance their segmentation performance on medical images, achieving an average accuracy increase of approximately 5 percentage points across validation sets.

What carries the argument

The Implantable Adaptive Cell (IAC): a compact module identified via Partially-Connected DARTS and designed for insertion into skip connections to adapt the flow of features in a frozen U-Net.

If this is right

  • Pre-trained U-Net models can receive performance upgrades through targeted module injection.
  • Segmentation accuracy improves by about 5 percentage points on average across multiple medical datasets.
  • Full retraining of complex networks is avoided, offering a lower-cost path to better results.
  • The method shows potential for use with other network architectures and tasks.

Where Pith is reading between the lines

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

  • Similar injection strategies might work on other parts of the network beyond skip connections.
  • Validating the method on datasets outside medical imaging would test its generality.
  • Determining if the gains depend on the specific DARTS search or could arise from any small module insertion would clarify the role of the architecture search.
  • Applying IACs to larger or more recent segmentation models could extend the observed benefits.

Load-bearing premise

The performance gains are caused by the specific modules found by the DARTS search rather than by the act of modifying the skip connections or other aspects of the experimental procedure.

What would settle it

An experiment that inserts randomly initialized or randomly structured small modules of the same size into the skip connections and measures whether comparable accuracy gains occur; absence of similar gains would support the claim that the searched cells are responsible.

Figures

Figures reproduced from arXiv: 2405.03420 by Emil Benedykciuk, Grzegorz W\'ojcik, Marcin Denkowski.

Figure 1
Figure 1. Figure 1: Diagram of the U-Net with Implantable Adaptive Cells (IAC) integration. It is worth noting that the BACKBONE is intentionally varied. Adaptive OP varies by stage: concat (Stage I), continuous IAC (Stage II), discrete IAC (Stage III). Detailed diagram of IAC is shown in [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: IAC diagram: input1 are coarse features procured via skip connections from the encoder, and input0 are upsampled decoder features. Search explores node connections and operations. A visual representation of the node block (yellow) is shown in [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 4
Figure 4. Figure 4: α and β computing and Partial Channel Connection ψ0,3 Partial Channel Connection Node 0 ψ1,3 Partial Channel Connection Node 1 ψ2,3 Partial Channel Connection Node 2 + Node 3 [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 3
Figure 3. Figure 3: A visual representation of the process of searching for node connections that minimize the uncertainty provoked by sampling (refer to [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 5
Figure 5. Figure 5: The base U-Net architectures were trained utilizing train dt and val dt sets, effectively serving as reference points in our studies. These datasets were also used in the training of the networks to which the IAC cell had been integrated. Two equally numbered sets, train search dt and val search dt, exclusively derived from the train dt set, were used in uncovering the adaptive cell’s architecture. B. Adap… view at source ↗
Figure 6
Figure 6. Figure 6: Example learning curve graphs for selected architectures with discretized cells obtained during training in Stage III. For each architecture, 5 cell genotypes were selected from epochs in Stage II. In each graph, curves have been drawn for the train dt (TRAIN CELL, in red–yellow color) and val dt (VAL CELL, in bluish color) datasets. of U-Net++. Paired t-tests and Wilcoxon tests confirm that improvements a… view at source ↗
read the original abstract

This paper introduces a novel approach to enhance the performance of pre-trained neural networks in medical image segmentation using gradient-based Neural Architecture Search (NAS) methods. We present the concept of Implantable Adaptive Cell (IAC), small modules identified through Partially-Connected DARTS based approach, designed to be injected into the skip connections of an existing and already trained U-shaped model. Unlike traditional NAS methods, our approach refines existing architectures without full retraining. Experiments on four medical datasets with MRI and CT images show consistent accuracy improvements on various U-Net configurations, with segmentation accuracy gain by approximately 5 percentage points across all validation datasets, with improvements reaching up to 11\%pt in the best-performing cases. The findings of this study not only offer a cost-effective alternative to the complete overhaul of complex models for performance upgrades but also indicate the potential applicability of our method to other architectures and problem domains.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 0 minor

Summary. The paper introduces Implantable Adaptive Cells (IACs), small modules discovered via a Partially-Connected DARTS approach, that are injected into the skip connections of already-trained U-Net models for medical image segmentation. The method is presented as refining existing architectures without full retraining. Experiments on four MRI and CT medical datasets report consistent Dice score improvements of approximately 5 percentage points across validation sets, with gains up to 11 points on various U-Net configurations.

Significance. If verified, the ability to achieve substantial Dice gains by training only the inserted IAC modules on frozen pre-trained U-Nets would constitute a meaningful efficiency advance for medical segmentation, providing a low-cost upgrade path and potential extension to other architectures. The manuscript currently supplies no evidence that the reported gains satisfy the frozen-base condition or arise specifically from the NAS-derived cell structure rather than added capacity.

major comments (2)
  1. [Abstract] Abstract: the central claim of 'consistent accuracy improvements ... without full retraining' and 'segmentation accuracy gain by approximately 5 percentage points' rests on experimental outcomes, yet the abstract (and by extension the reported results) supplies no baselines, statistical tests, error bars, dataset splits, or controls for confounding variables, making it impossible to determine whether the data support the claim.
  2. [Experiments] Experiments section: the claim that Partially-Connected DARTS cells can be inserted into skip connections of a frozen pre-trained U-Net and produce the stated gains solely by optimizing the new modules lacks any protocol confirming that encoder/decoder parameters remain fixed, and contains no ablations (random cells, zero cells, or parameter-matched additions) that would isolate the NAS-derived structure as the cause.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. The comments highlight important aspects of experimental rigor that we will address in revision.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim of 'consistent accuracy improvements ... without full retraining' and 'segmentation accuracy gain by approximately 5 percentage points' rests on experimental outcomes, yet the abstract (and by extension the reported results) supplies no baselines, statistical tests, error bars, dataset splits, or controls for confounding variables, making it impossible to determine whether the data support the claim.

    Authors: We agree the abstract is concise and omits explicit references to these elements. The main text and supplementary material contain the requested details (dataset splits, baseline Dice scores, and per-fold results). In revision we will expand the abstract slightly to note the use of four public datasets, frozen-base training, and consistent gains relative to the unmodified U-Net baselines. revision: partial

  2. Referee: [Experiments] Experiments section: the claim that Partially-Connected DARTS cells can be inserted into skip connections of a frozen pre-trained U-Net and produce the stated gains solely by optimizing the new modules lacks any protocol confirming that encoder/decoder parameters remain fixed, and contains no ablations (random cells, zero cells, or parameter-matched additions) that would isolate the NAS-derived structure as the cause.

    Authors: This observation is correct; the current manuscript states that the base U-Net is frozen but does not supply an explicit training protocol or the suggested ablations. We will add a dedicated subsection describing the freezing procedure (optimizer applied only to IAC parameters, learning-rate schedule, and early-stopping criteria) together with ablation experiments that replace IACs with random cells and with parameter-matched linear layers to isolate the contribution of the discovered cell topology. revision: yes

Circularity Check

0 steps flagged

No significant circularity; experimental claims rest on reported dataset outcomes rather than definitional reduction.

full rationale

The paper presents an empirical method: Partially-Connected DARTS is used to identify IAC modules that are then inserted into skip connections of already-trained U-Nets. Performance gains (5–11 pt Dice) are asserted from experiments on four medical datasets. No equations, predictions, or first-principles derivations appear in the abstract or described approach that reduce to the inputs by construction. No self-citations are invoked as load-bearing uniqueness theorems, and the NAS component is a standard external technique. The central claim therefore remains falsifiable via the reported validation metrics and does not collapse into a tautology or fitted renaming.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 1 invented entities

Based solely on the abstract, the central claim rests on the unverified effectiveness of IAC modules discovered via DARTS and their compatibility with pre-trained U-Net skip connections; no free parameters, axioms, or invented entities can be audited in detail.

invented entities (1)
  • Implantable Adaptive Cell (IAC) no independent evidence
    purpose: Small neural module to be inserted into U-Net skip connections for performance enhancement without full retraining
    New concept introduced in the abstract to achieve the reported accuracy gains.

pith-pipeline@v0.9.0 · 5691 in / 1180 out tokens · 31290 ms · 2026-05-24T01:40:00.311192+00:00 · methodology

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

43 extracted references · 43 canonical work pages · 4 internal anchors

  1. [1]

    Systematic review on neural architecture search,

    S. Salmani Pour Avval, N. D. Eskue, R. M. Groves, and V . Y ag houbi, “Systematic review on neural architecture search,” Artificial Intelligence Review, vol. 58, no. 3, Jan. 2025

    work page 2025

  2. [2]

    Auto-DeepLab: Hierarchical Neural Architect ure Search for Semantic Image Segmentation,

    C. Liu, L.-C. Chen, F. Schroff, H. Adam, W. Hua, A. L. Y uill e, and L. Fei-Fei, “Auto-DeepLab: Hierarchical Neural Architect ure Search for Semantic Image Segmentation,” in 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , 2019, pp. 82–92

    work page 2019

  3. [3]

    Fast erSeg: Searching for Faster Real-time Semantic Segmentation,

    W. Chen, X. Gong, X. Liu, Q. Zhang, Y . Li, and Z. Wang, “Fast erSeg: Searching for Faster Real-time Semantic Segmentation,” in 8th Inter- national Conference on Learning Representations, ICLR 202 0, Addis Ababa, Ethiopia, April 26-30, 2020 . OpenReview.net, 2020

    work page 2020

  4. [4]

    DCNAS: Densely Connected Neural Architecture Search for S emantic Image Segmentation,

    X. Zhang, H. Xu, H. Mo, J. Tan, C. Y ang, L. Wang, and W. Ren, “DCNAS: Densely Connected Neural Architecture Search for S emantic Image Segmentation,” in IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2021, virtual, June 19-25, 2021 . Computer Vision Foundation / IEEE, 2021, pp. 13 956–13 967

    work page 2021

  5. [5]

    Automatically Designing CNN Ar chitectures for Medical Image Segmentation,

    A. Mortazi and U. Bagci, “Automatically Designing CNN Ar chitectures for Medical Image Segmentation,” in Machine Learning in Medical Imaging - 9th International W orkshop, MLMI 2018, Held in Con junction with MICCAI 2018, Granada, Spain, September 16, 2018, Proce edings, ser. Lecture Notes in Computer Science, Y . Shi, H. Suk, and M. Liu, Eds., vol. 11046...

    work page 2018

  6. [6]

    NAS-Unet: Neural Arch itecture Search for Medical Image Segmentation,

    Y . Weng, T. Zhou, Y . Li, and X. Qiu, “NAS-Unet: Neural Arch itecture Search for Medical Image Segmentation,” IEEE Access , vol. 7, pp. 44 247–44 257, 2019

    work page 2019

  7. [7]

    Re- source Optimized Neural Architecture Search for 3D Medical Image Segmentation,

    W. Bae, S. Lee, Y . Lee, B. Park, M. Chung, and K. Jung, “Re- source Optimized Neural Architecture Search for 3D Medical Image Segmentation,” in Medical Image Computing and Computer Assisted Intervention - MICCAI 2019 - 22nd International Conference , Shenzhen, China, October 13-17, 2019, Proceedings, Part II , ser. Lecture Notes in Computer Science, D. S...

    work page 2019

  8. [8]

    V -NAS: Neura l Architecture Search for V olumetric Medical Image Segmenta tion,

    Z. Zhu, C. Liu, D. Y ang, A. Y uille, and D. Xu, “V -NAS: Neura l Architecture Search for V olumetric Medical Image Segmenta tion,” in 2019 International Conference on 3D Vision (3DV) , 2019, pp. 240–248

    work page 2019

  9. [9]

    DiNTS: Differe ntiable Neural Network Topology Search for 3D Medical Image Segment ation,

    Y . He, D. Y ang, H. Roth, C. Zhao, and D. Xu, “DiNTS: Differe ntiable Neural Network Topology Search for 3D Medical Image Segment ation,” in IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2021, virtual, June 19-25, 2021 . Computer Vision Foundation / IEEE, 2021, pp. 5841–5850

    work page 2021

  10. [10]

    HyperSegNAS: Bridgi ng One-Shot Neural Architecture Search with 3D Medical Image S egmen- tation using HyperNet,

    C. Peng, A. Myronenko, A. Hatamizadeh, V . Nath, M. M. R. S iddiquee, Y . He, D. Xu, R. Chellappa, and D. Y ang, “HyperSegNAS: Bridgi ng One-Shot Neural Architecture Search with 3D Medical Image S egmen- tation using HyperNet,” in IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2022, New Orleans, LA, USA, Ju ne 18-24, 2022 . IEEE, 202...

    work page 2022

  11. [11]

    DARTS: Differentiabl e Architecture Search,

    H. Liu, K. Simonyan, and Y . Y ang, “DARTS: Differentiabl e Architecture Search,” in 7th International Conference on Learning Representations, ICLR 2019, New Orleans, LA, USA, May 6-9, 2019 . OpenReview.net, 2019. [Online]. Available: https://openreview.net/forum?id=S1eYHoC5FX

    work page 2019

  12. [12]

    DARTS+: Improved Differentiable Architecture Search wit h Early Stopping,

    H. Liang, S. Zhang, J. Sun, X. He, W. Huang, K. Zhuang, and Z. Li, “DARTS+: Improved Differentiable Architecture Search wit h Early Stopping,” CoRR, vol. abs/1909.06035, 2019

    work page arXiv 1909

  13. [13]

    PC- DARTS: Partial Channel Connections for Memory-Efficient Ar chitecture Search,

    Y . Xu, L. Xie, X. Zhang, X. Chen, G. Qi, Q. Tian, and H. Xion g, “PC- DARTS: Partial Channel Connections for Memory-Efficient Ar chitecture Search,” in 8th International Conference on Learning Representations , ICLR 2020, Addis Ababa, Ethiopia, April 26-30, 2020 . OpenReview.net, 2020

    work page 2020

  14. [14]

    DropNAS: Grouped Operation Dropout for Differentiable Ar chitecture Search,

    W. Hong, G. Li, W. Zhang, R. Tang, Y . Wang, Z. Li, and Y . Y u, “DropNAS: Grouped Operation Dropout for Differentiable Ar chitecture Search,” CoRR, vol. abs/2201.11679, 2022

    work page arXiv 2022

  15. [15]

    Single-DARTS: Towards Stab le Archi- 10 IEEE TRANSACTIONS ON MEDICAL IMAGING, VOL. XX, NO. XX, XXX X 2020 tecture Search,

    P . Hou, Y . Jin, and Y . Chen, “Single-DARTS: Towards Stab le Archi- 10 IEEE TRANSACTIONS ON MEDICAL IMAGING, VOL. XX, NO. XX, XXX X 2020 tecture Search,” in IEEE/CVF International Conference on Computer Vision W orkshops, ICCVW 2021, Montreal, BC, Canada, Octobe r 11- 17, 2021 . IEEE, 2021, pp. 373–382

    work page 2020

  16. [16]

    Deep learning techniques for automatic mri cardiac multi- structures segmentation and diagnosis: Is the problem solved?

    O. Bernard, A. Lalande, C. Zotti, F. Cervenansky, X. Y an g, P .-A. Heng, I. Cetin, K. Lekadir, O. Camara, M. A. Gonzalez Ballest er, G. Sanroma, S. Napel, S. Petersen, G. Tziritas, E. Grinias, M . Khened, V . A. Kollerathu, G. Krishnamurthi, M.-M. Roh´ e, X. Pennec, M. Ser- mesant, F. Isensee, P . J¨ ager, K. H. Maier-Hein, P . M. Full, I . Wolf, S. Enge...

    work page 2018

  17. [17]

    The Multimodal Brain Tumor Image Segmentation Benchmark (BRA T S),

    B. H. Menze, A. Jakab, S. Bauer, J. Kalpathy-Cramer, K. F arahani, J. Kirby, Y . Burren, N. Porz, J. Slotboom, R. Wiest et al. , “The Multimodal Brain Tumor Image Segmentation Benchmark (BRA T S),” IEEE Transactions on Medical Imaging , vol. 34, no. 10, pp. 1993–2024, Oct 2015

    work page 1993

  18. [18]

    Segmentation labels and ra diomic features for the pre-operative scans of the tcga-gbm collec tion,

    S. Bakas, A. Sotiras, M. Bilello, M. Rozycki, J. Kirby, J . Freymann, K. Farahani, and C. Davatzikos, “Segmentation labels and ra diomic features for the pre-operative scans of the tcga-gbm collec tion,” 07 2017

    work page 2017

  19. [19]

    The Brain Tumor Segmentation (Br aTS) Challenge 2023: Brain MR Image Synthesis for Tumor Segmenta tion (BraSyn),

    H. B. Li, G. M. Conte, S. M. Anwar, F. Kofler, K. V . Leemput, M. Piraud, I. Ezhov, F. Meissen, M. Adewole, A. Janas, A. F. Ka zerooni, D. LaBella, A. W. Moawad, K. Farahani, J. A. Eddy, T. Bergquis t, V . Chung, R. T. Shinohara, F. Dako, W. I. Wiggins, Z. Reitman,C. Wang, X. Liu, Z. Jiang, A. Familiar, E. Johanson, Z. Meier, C. Davat zikos, J. B. Freymann...

    work page arXiv 2023

  20. [20]

    The KiTS21 Challenge: Automatic segmentation of kidneys, renal tumors, and renal cysts in corticomedullary-phase ct,

    N. Heller, F. Isensee, D. Trofimova, R. Tejpaul, Z. Zhao, H. Chen, L. Wang, A. Golts, D. Khapun, D. Shats, Y . Shoshan, F. Gilboa- Solomon, Y . George, X. Y ang, J. Zhang, J. Zhang, Y . Xia, M. Wu, Z. Liu, E. Walczak, S. McSweeney, R. V asdev, C. Hornung, R. Solaiman, J. Schoephoerster, B. Abernathy, D. Wu, S. Abdu lkadir, B. Byun, J. Spriggs, G. Struyk, A...

    work page arXiv

  21. [21]

    Amos: a large-scale abdominal multi-organ benchmark for versatile medical image segmentation,

    Y . Ji, H. Bai, C. Ge, J. Y ang, Y . Zhu, R. Zhang, Z. Li, L. Zha ng, W. Ma, X. Wan, and P . Luo, “Amos: a large-scale abdominal multi-organ benchmark for versatile medical image segmentation,” in Proceedings of the 36th International Conference on Neural Information Processing Systems, ser. NIPS ’22. Red Hook, NY , USA: Curran Associates Inc., 2024

    work page 2024

  22. [22]

    Ronneberger, P

    O. Ronneberger, P . Fischer, and T. Brox, U-Net: Convolutional Networks for Biomedical Image Segmentation , ser. Lecture Notes in Computer Science. Springer, 2015, vol. 9351, pp. 234–241

    work page 2015

  23. [23]

    Attention U-Net: Learning Where to Look for the Pancreas

    O. Oktay, J. Schlemper, L. L. Folgoc, M. C. H. Lee, M. P . He inrich, K. Misawa, K. Mori, S. G. McDonagh, N. Y . Hammerla, B. Kainz, B. Glocker, and D. Rueckert, “Attention U-Net: Learning Whe re to Look for the Pancreas,” CoRR, vol. abs/1804.03999, 2018

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  24. [24]

    RAUNet: Residual Attention U-Net for Semantic Segme ntation of Cataract Surgical Instruments,

    Z. Ni, G. Bian, X. Zhou, Z. Hou, X. Xie, C. Wang, Y . Zhou, R. Li, and Z. Li, “RAUNet: Residual Attention U-Net for Semantic Segme ntation of Cataract Surgical Instruments,” in Neural Information Processing - 26th International Conference, ICONIP 2019, Sydney, NSW , A ustralia, December 12-15, 2019, Proceedings, Part II , ser. Lecture Notes in Computer Sc...

    work page 2019

  25. [25]

    UNet++: Redesigning Skip Connections to Exploit Multiscale Featur es in Image Segmentation,

    Z. Zhou, M. M. R. Siddiquee, N. Tajbakhsh, and J. Liang, “ UNet++: Redesigning Skip Connections to Exploit Multiscale Featur es in Image Segmentation,” IEEE Transactions on Medical Imaging , vol. 39, no. 6, pp. 1856–1867, 2020

    work page 2020

  26. [26]

    nnU-Net: Breaking the Spell on Successful Medical Im age Segmentation,

    F. Isensee, J. Petersen, S. A. A. Kohl, P . F. J¨ ager, and K . H. Maier- Hein, “nnU-Net: Breaking the Spell on Successful Medical Im age Segmentation,” CoRR, vol. abs/1904.08128, 2019

    work page arXiv 1904

  27. [27]

    nnU-Net Revisited: A Call for Rigoro us V alida- tion in 3D Medical Image Segmentation

    F. Isensee, T. Wald, C. Ulrich, M. Baumgartner, S. Roy, K . H. Maier- Hein, and P . F. J¨ ager, “nnU-Net Revisited: A Call for Rigoro us V alida- tion in 3D Medical Image Segmentation.” in MICCAI (9) , ser. Lecture Notes in Computer Science, M. G. Linguraru, Q. Dou, A. Ferage n, S. Giannarou, B. Glocker, K. Lekadir, and J. A. Schnabel, Eds ., vol. 15009. ...

    work page 2024

  28. [28]

    A large annotated medical image dataset for the development and evaluation of segmentation algorithms

    A. L. Simpson, M. Antonelli, S. Bakas, M. Bilello, K. Far ahani, B. van Ginneken, A. Kopp-Schneider, B. A. Landman, G. Litjens, B. H . Menze, O. Ronneberger, R. M. Summers, P . Bilic, P . F. Christ, R. K. G. Do, M. Gollub, J. Golia-Pernicka, S. Heckers, W. R. Jarnagin, M. McHugo, S. Napel, E. V orontsov, L. Maier-Hein, and M. J. Cardoso, “A l arge annotat...

    work page internal anchor Pith review Pith/arXiv arXiv 1902

  29. [29]

    Neural Architecture Search with Reinforcement Learning

    B. Zoph and Q. V . Le, “Neural Architecture Search with Re inforcement Learning,” CoRR, vol. abs/1611.01578, 2016

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  30. [30]

    Large-Scale Evolution of Image Classifi ers,

    E. Real, S. Moore, A. Selle, S. Saxena, Y . I. Leon-Suemat su, J. Tan, Q. V . Le, and A. Kurakin, “Large-Scale Evolution of Image Classifi ers,” in Proceedings of the 34th International Conference on Machin e Learning, ICML 2017, Sydney, NSW , Australia, 6-11 August 2017, ser. Proceedings of Machine Learning Research, D. Precup and Y . W. Teh, Eds., v ol. ...

    work page 2017

  31. [31]

    Progressive Differe ntiable Architecture Search: Bridging the Depth Gap Between Search and Evaluation,

    X. Chen, L. Xie, J. Wu, and Q. Tian, “Progressive Differe ntiable Architecture Search: Bridging the Depth Gap Between Search and Evaluation,” in 2019 IEEE/CVF International Conference on Computer Vision (ICCV), 2019, pp. 1294–1303

    work page 2019

  32. [32]

    ℓ-darts: Light-weight differentiable architecture search with robustness enhancement strategy,

    L. Hu, Z. Wang, H. Li, P . Wu, J. Mao, and N. Zeng, “ ℓ-darts: Light-weight differentiable architecture search with robustness enhancement strategy,” Knowledge-Based Systems, vol. 288, p. 111466, 2024. [Online]. Available: https://www.sciencedirect.com/science/article/pii/S0950705124001011

    work page 2024

  33. [33]

    Medical neural architecture search: Survey and taxonomy,

    H. Benmeziane, I. Hamzaoui, Z. Cherif, and K. El Maghrao ui, “Medical neural architecture search: Survey and taxonomy,” in Proceedings of the Thirty-Third International Joint Conference on Artificial Intelligence, IJCAI-24, K. Larson, Ed. International Joint Conferences on Artifici al Intelligence Organization, 8 2024, pp. 7932–7940, survey T rack. [Online]...

    work page doi:10.24963/ijcai.2024/878 2024

  34. [34]

    Neural archit ecture search for biomedical image classification: A comparative study ac ross data modalities,

    Z. Kus ¸, M. Aydin, B. Kiraz, and A. Kiraz, “Neural archit ecture search for biomedical image classification: A comparative study ac ross data modalities,” Artificial Intelligence in Medicine , vol. 160, p. 103064, 2025

    work page 2025

  35. [35]

    The medical segmentation decathlon,

    M. Antonelli, A. Reinke, S. Bakas, K. Farahani, A. Kopp- Schneider, B. A. Landman, G. Litjens, B. Menze, O. Ronneberger, R. M. Sum mers, B. van Ginneken, M. Bilello, P . Bilic, P . F. Christ, R. K. G. Do , M. J. Gollub, S. H. Heckers, H. Huisman, W. R. Jarnagin, M. K. McHug o, S. Napel, J. S. G. Pernicka, K. Rhode, C. Tobon-Gomez, E. V oro ntsov, J. A. Me...

    work page 2022

  36. [36]

    EfficientNetV2: Smaller models and fas ter training,

    M. Tan and Q. Le, “EfficientNetV2: Smaller models and fas ter training,” in Proceedings of the 38th International Conference on Machin e Learning, ser. Proceedings of Machine Learning Research, M. Meila an d T. Zhang, Eds., vol. 139. PMLR, 18–24 Jul 2021, pp. 10 096–10 1 06. [Online]. Available: https://proceedings.mlr.press/v139/tan21a.html

    work page 2021

  37. [37]

    A ConvNet for the 2020s,

    Z. Liu, H. Mao, C.-Y . Wu, C. Feichtenhofer, T. Darrell, a nd S. Xie, “A ConvNet for the 2020s,” in 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , 2022, pp. 11 966–11 976

    work page 2022

  38. [38]

    Searching for MobileNetV3,

    A. Howard, M. Sandler, B. Chen, W. Wang, L.-C. Chen, M. Ta n, G. Chu, V . V asudevan, Y . Zhu, R. Pang, H. Adam, and Q. Le, “Searching for MobileNetV3,” in 2019 IEEE/CVF International Conference on Computer Vision (ICCV) , 2019, pp. 1314–1324

    work page 2019

  39. [39]

    MedNeXt: Transformer-d riven scaling of convnets for medical image segmentation,

    S. Roy, G. Koehler, C. Ulrich, M. Baumgartner, J. Peters en, F. Isensee, P . F. J¨ ager, and K. H. Maier-Hein, “MedNeXt: Transformer-d riven scaling of convnets for medical image segmentation,” in Medical Im- age Computing and Computer Assisted Intervention – MICCAI 2 023, H. Greenspan, A. Madabhushi, P . Mousavi, S. Salcudean, J. Du ncan, T. Syeda-Mahmoo...

    work page 2023

  40. [40]

    sharpDARTS: Faster and More Accurate Differentiable Architecture Search

    A. Hundt, V . Jain, and G. D. Hager, “sharpdarts: Faster a nd more accurate differentiable architecture search,” CoRR, vol. abs/1903.09900, 2019

    work page internal anchor Pith review Pith/arXiv arXiv 1903

  41. [41]

    β -darts: Beta- decay regularization for differentiable architecture sea rch,

    P . Y e, B. Li, Y . Li, T. Chen, J. Fan, and W. Ouyang, “ β -darts: Beta- decay regularization for differentiable architecture sea rch,” in IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVP R 2022, New Orleans, LA, USA, June 18-24, 2022 . IEEE, 2022, pp. 10 864– 10 873

    work page 2022

  42. [42]

    Swin-Unet: Unet-like pure transformer for medical image s egmenta- tion

    H. Cao, Y . Wang, J. Chen, D. Jiang, X. Zhang, Q. Tian, and M . Wang, “Swin-Unet: Unet-like pure transformer for medical image s egmenta- tion.” Springer Nature Switzerland, 2023, p. 205–218

    work page 2023

  43. [43]

    UNETR: Transformers for 3D Medica l Image Segmentation,

    A. Hatamizadeh, Y . Tang, V . Nath, D. Y ang, A. Myronenko, B. Land- man, H. R. Roth, and D. Xu, “UNETR: Transformers for 3D Medica l Image Segmentation,” in 2022 IEEE/CVF Winter Conference on Appli- cations of Computer Vision (WACV) . Los Alamitos, CA, USA: IEEE Computer Society, Jan. 2022, pp. 1748–1758

    work page 2022