Pith. sign in

REVIEW 3 major objections 8 minor 36 references

Distribution-aware Fairness Learning in Medical Image Segmentation From A Control-Theoretic Perspective

T0 review · 3 major / 8 minor · reviewed 2026-08-09 · deepseek-v4-flash

Pith's one-line read Conditioning sparse expert routing on demographic or clinical attributes improves segmentation fairness in medical imaging.

desk verdict A reasonable plug-in fairness mechanism whose empirical edge over its own MoE baseline is not statistically established, and whose control-theoretic derivation rests on an assumption the implemented experts violate. read the letter →

arxiv 2502.00619 v2 pith:YNZX6DOV submitted 2025-02-02 eess.IV cs.AIcs.CV

classification eess.IVcs.AIcs.CV
keywords medicalimagesegmentationfairnesslearningmixtureofexpertssparsegatingoptimalcontroltheoryequity-scaledDicedemographicattributesclinical
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

The paper's central claim is that segmentation models trained on imbalanced clinical data can be made fairer by conditioning a sparse mixture-of-experts gating network on an attribute flag, such as race, age, or tumor stage. The proposed dMoE module keeps a set of shared expert networks and learns a separate top-k router for each attribute value, so that underrepresented subgroups effectively switch among specialized subnetworks while still sharing what they can. On the FairSeg, HAM10000, and an in-house prostate radiotherapy CT dataset, dMoE reports the best equity-scaled Dice and IoU among the compared fairness methods, with the clearest gains in the least-represented subgroups. The paper also interprets this design as mode-switching control within an optimal-control view of neural network training, arguing that external distributional information acts as a switch among control modes. If the empirical claims hold, the method offers a practical way to turn available demographic and clinical labels into measurable fairness improvements in medical image segmentation.

What carries the argument

The load-bearing object is the dMoE layer itself, defined by Eqs. (2)-(5): a sparse gating network $\mathbf{G}_{attr}$ that takes the flattened hidden state and an attribute flag, injects Gaussian noise, applies KeepTop-$k$, and forms a weighted mixture of shared expert MLPs whose output is added as a residual to the hidden state. The control-theoretic bridge is the identity in Eq. (16), which rewrites a kernel-weighted sum of expert parameters as a mixture of expert outputs when the layer transformation is linear in its second argument; this identity connects the kernel approximation of the control policy to the MoE gating weights, and the attribute-specific matrices in Eq. (18) implement mode-switching. The paper's derivation is conditional on that linearity, while the implemented experts are two-layer ReLU MLPs, so the identity holds conceptually rather than numerically for the trained architecture.

What would settle it

Evaluate both sides of Eq. (16) on real hidden states from a trained dMoE layer; any difference beyond floating-point noise shows that the derivation's key equality fails for the implemented experts. Alternatively, replace the ReLU expert MLPs with linear experts and retrain: if the equity-scaled gains disappear, the nonlinear experts rather than the mode-switching control design are carrying the improvement.

Watch

Extended reading notes

Core claim

dMoE is a plug-in layer for segmentation networks: after a layer block, the hidden embeddings are flattened, an attribute-specific router computes noisy top-k weights over eight shared expert MLPs, and the weighted expert outputs are added back as a residual. Each attribute value (race group, age band, T-stage) trains its own router, while the experts are shared across the whole population, which the authors argue lets minority subgroups benefit from the model capacity without losing the common representation. Across three datasets and two backbone families, the paper reports consistent improvements in ES-Dice and ES-IoU relative to plain MoE, FEBS, DRO, adversarial training, and generative augmentation, including a rise in ES-Dice from 0.733 with MoE to 0.743 with dMoE on FairSeg rim segmentation, from 0.796 to 0.801 on HAM10000, and from 0.452 to 0.499 on the prostate CT dataset. The authors read this result through control theory: the layer-wise hidden-state update is a discrete dynamical system, standard MoE is a feedback controller, and dMoE is a mode-switching controller whose mode is selected by the attribute flag.

Load-bearing premise

The argument assumes that averaging expert parameters and then applying the layer gives the same result as applying the layer to each expert and averaging the outputs, but the implemented experts are nonlinear ReLU networks, so this equality does not hold for the actual model.

Editorial extensions

If this is right

  • If dMoE's results are correct, attribute-conditioned routing can be added to existing transformer and CNN segmentation backbones without changing the loss function, converting demographic or clinical labels already present in many clinical datasets into fairness gains.
  • The reported experiments imply that fairness in medical segmentation can be improved by architectural conditioning on known attributes, not only by loss reweighting, adversarial debiasing, or generative augmentation.
  • The success on T-stage, Gleason grade, and PSA-level attributes suggests the same mechanism can address clinical factors that are not directly visible in the image, such as disease severity and treatment intent.
  • The ablation studies attribute the improvement specifically to the mode-switching (attribute-aware) control component rather than to the added MoE capacity alone.

Reading between the lines

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

  • Beyond the paper: because Eq. (16) requires linearity that the ReLU experts violate, the reported gains, not the control-theoretic derivation, are what establish dMoE's value; testing the same routing with linear experts would separate the two.
  • Beyond the paper: the model uses the attribute flag at inference time, so a natural stress test is to withhold or corrupt the attribute at deployment and measure whether fairness degrades, which the paper does not report.
  • Beyond the paper: the same gating pattern could transfer to other dense prediction tasks and to multi-attribute settings, for example age and stage jointly, where a hierarchical or composed router might handle combined imbalances.
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 / 8 minor

Summary. The paper proposes Distribution-aware Mixture of Experts (dMoE), a sparse-gating layer whose router is conditioned on a demographic or clinical attribute flag, and inserts it into the encoders of segmentation backbones (TransUNet, 3D ResUNet). The authors reinterpret standard MoE as feedback control and dMoE as attribute-wise mode-switching control (Section 3.3), then evaluate on Harvard-FairSeg (race), HAM10000 (age), and an in-house prostate CT dataset (T-stage, Gleason Grade Group, PSA), reporting equity-scaled Dice/IoU (ES-Dice/ES-IoU) together with subgroup Dice/IoU and comparisons against ADV, DRO, FEBS, FairDiff, and a default MoE baseline. The headline claims are SOTA ES-Dice values of 0.743 on the FairSeg rim task, 0.801 on HAM10000, and 0.499 on the prostate CT dataset, with particularly large subgroup gains reported for T1 (n=11) and T4 (n=21) prostate cases.

Significance. The empirical program is substantive: evaluation spans 2D and 3D segmentation, demographic and clinical attributes, and out-of-center test cohorts (Yongin and Gangnam Severance), and the main comparisons are benchmarked, so the fairness results are not artifacts of the control-theoretic framing. The released code, the ablations over module placement and parameter sharing, and the extension to clinical attributes (T-stage, GG, PSA) beyond demographics are genuine strengths. If the empirical gains hold up under proper paired inference, dMoE is a practical, architecture-agnostic mechanism with modest overhead. However, three load-bearing issues currently separate the claims from the evidence: the theoretical derivation in Section 3.3 relies on conditions that the implemented architecture violates; the decisive dMoE-versus-MoE comparison is not statistically supported as reported; and the appendix ablation tables are internally inconsistent with the main tables. The control-theoretic narrative is an analogy rather than a circularity, but the paper presents it as a derivation.

major comments (3)
  1. [Section 3.3, Eqs. (13)-(16); Section 4.2] The derivation in Eqs. (13)-(16) identifies dMoE with mode-switching optimal control only if the transformation f is linear in its second argument and if the gating is a dense kernel-weighted sum over all experts; neither condition holds for the implemented module. Section 4.2 specifies each expert as a two-layer MLP with ReLU and dropout, which is nonlinear in its parameters, and Eqs. (3)-(5) with Top-k=2 implement sparse top-2 routing, whereas Eq. (16) uses the dense kernel weights G_i = K(h_t, h^i_t) summed over all i. The equality in Eq. (16) therefore does not hold for the actual dMoE layer, so the sentence "This formulation naturally leads us to mode-switching control" overstates what has been shown; Eqs. (10)-(11) are a restatement of training as a terminal-cost problem, with no optimal-control result (e.g., necessary conditions) derived or used. I recommend reframing Section 3.3 explicitly as an analogy or heuristic motivation, or providing a version of Eq. (16) that holds under the actual top-k routing and nonlinear experts, and tempering the abstract's claim of a "comprehensive analysis of its underlying mechanisms." This issue does not by itself invalidate the empirical comparisons, which are benchmarked independently of the framing.
  2. [Section 4.3, Section 4.4.2, Tables 1-3] The comparison that isolates the paper's core novelty, dMoE versus the default MoE ablation, is not statistically supported as reported. The bootstrapped 95% CIs overlap in every dataset at the All level: rim ES-Dice 0.743 (0.723-0.763) versus 0.733 (0.713-0.752); cup ES-Dice 0.832 (0.810-0.853) versus 0.830 (0.809-0.847); HAM ES-Dice 0.801 (0.745-0.847) versus 0.796 (0.741-0.844); prostate ES-Dice 0.499 (0.469-0.531) versus 0.452 (0.415-0.492). No paired difference interval or test is reported anywhere, even though a paired bootstrap over per-subject Dice differences would be the natural procedure given the resampling setup described in Section 4.3. The subgroup headline numbers (T1 Dice 0.718 with n=11, T4 Dice 0.778 with n=21, Age<20 Dice 0.880 with n=24) are given without subgroup CIs, and Section 4.4.2 calls the T4 result "significantly outperforming" without any test. The manuscript should either report paired inference for dMoE versus MoE (for example, a bootstrap CI on the per-case difference), add subgroup CIs and justify the use of "significantly," or temper the SOTA claim to point estimates whose uncertainty is acknowledged.
  3. [Appendix A.5, Tables 11-12 versus Tables 2-3] The ablation tables are internally inconsistent with the main results and with one another. For HAM10000, the configuration labeled "Ours" in Table 11 (encoder-only, layer-wise, mode-switching) reports ES-Dice 0.841, whereas the same configuration in Table 2 reports ES-Dice 0.801, even though the All Dice values agree at 0.884. For the prostate dataset, Table 12 contains three "Ours" rows describing the same configuration with different values: ES-Dice 0.546, 0.546, and 0.499 with All Dice 0.711, 0.710, and 0.650 across the Location, Parameters, and Optimal-Control subsections, while Table 3 reports 0.499 and 0.650 for that configuration. In addition, Table 12 shows non-feedback control with ES-Dice 0.509, which is higher than the mode-switching value of 0.499 on the same task; this contradicts the claim in Appendix A.5 that mode-switching control "consistently improved both ES-Dice and Dice scores" (Dice improves, 0.650 versus 0.615, but ES-Dice does not). The authors should reconcile these numbers, state which run is canonical, and qualify the consistency claim, since the ablation currently does not support the stated conclusion.
minor comments (8)
  1. [Section 3.2 and Eq. (5)] There are several wording and typesetting errors: "flattend" should be "flattened" in the paragraph before Eq. (2); the description after Eq. (5) says KeepTop-k "retrains only the top-k" but should say "retains"; Section 4.4.1 contains "dMoe" instead of "dMoE"; Section 4.4.3 contains "These finding show"; the Figure 3 caption says "Violin plots depicts" instead of "depict."
  2. [Eq. (6)] The segmentation loss in Eq. (6) is typeset incorrectly: the expectation operator has a malformed subscript, the loss function symbol is garbled, and since Eq. (7) defines the prediction as a function of both the image and the attribute flag, Eq. (6) should make the dependence on the attribute explicit rather than writing the loss only over X.
  3. [Abstract] The abstract contains two consecutive sentences announcing code availability ("The source code will be made available. The source code is available at https://github.com/tvseg/dMoE."); one of them should be deleted.
  4. [Section 4.2] The checkpoint-selection protocol is unclear: "The best performance is selected from checkpoints saved at 100-epoch intervals" is ambiguous for HAM10000, which is trained for only 100 epochs, and no validation split is mentioned for either 2D task. If "best performance" was evaluated on the test benchmark, the reported numbers would be optimistically biased; the selection criterion and the data used for selection should be stated explicitly (the 3D task describes early stopping on a validation set, so the 2D protocol should likewise be specified).
  5. [Table 6 and Table 2] The HAM10000 subgroup counts do not sum to the stated total: 121+469+328+120+24 = 1062, while the tables state n=1061. In addition, the age column labels "≥80," "≥60," "≥40," "≥20," and "<20" are cumulative thresholds, but the reported counts (e.g., 469 for "≥60" versus 328 for "≥40") show the groups are actually disjoint bins; the labels should be ranges such as "80+," "60-79," "40-59," "20-39," and "<20" to avoid confusion.
  6. [Table 1 footnotes] The FairDiff ES-metrics are "recalculated using Eq. (19) based on metrics reported in the original paper." Please describe how the per-subgroup Dice/IoU values required by Eq. (20) were obtained for FairDiff, and state whether the identical recalculation procedure was applied to the other externally reported baselines, so that the comparison is apples-to-apples.
  7. [References] The reference Langley (2000) appears in the bibliography, but I could not locate an in-text citation for it; it should be cited where relevant or removed.
  8. [Eqs. (12) and (17)] Equations (12) and (17) are written in differential form ("d h_t = ... dt") with no stochastic term, while the rest of the presentation is in terms of discrete layer updates and ODEs; the differential notation should be made consistent with the deterministic setting, or the stochastic convention should be introduced explicitly.

Circularity Check

0 steps flagged · score 2.0 of 10

No substantive circularity: the fairness results are externally benchmarked, and the control-theoretic framing is a post-hoc analogy rather than a fitted prediction; only minor non-load-bearing self-citations appear.

full rationale

The paper's core empirical claim—dMoE improves equity-scaled Dice/IoU over MoE, FEBS, and FairDiff on three datasets—is tested against external, held-out benchmarks (Section 4.4, Tables 1-3), and the network is trained solely with the segmentation loss (Eq. 6), not by optimizing the reported ES-Dice metric. The control-theoretic bridge in Section 3.3 (Eq. 16) is an interpretation: it rewrites a kernel-weighted control sum as MoE gating under an explicit linearity assumption on f. Because the implemented experts are described as two-layer MLPs with ReLU (Section 4.2), that equality does not hold for the actual architecture, so the 'mode-switching optimal control' claim is an analogy rather than a derivation. This is a validity gap, not a circular reduction: no result is forced by construction, and the empirical gains do not depend on the analogy. The only self-citations (Oh et al., 2024, to justify the 3D ResUNet backbone; Jin et al., 2024, for fairness motivation) are not load-bearing: the dMoE-vs-MoE comparison uses a standard MoE baseline and external methods such as FEBS and FairDiff. The overlapping confidence intervals on the dMoE-vs-MoE comparison (Tables 1-3) are a statistical-evidence concern, not circularity. The paper's own limitations section (Section 5) concedes that performance improvements vary by task and dataset, consistent with a post-hoc design rationale rather than a circular derivation. Overall, the central fairness claim is self-contained against external benchmarks, so the only deduction is a minor, non-load-bearing self-citation.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

No new physical or ontological entities are postulated; dMoE is a neural architecture module, not a new force or particle. The theoretical framing is an analogy, not a new conserved quantity.

free parameters (3)
  • Top-k (number of active experts) = 2
    Set to 2 for all experiments without reported sensitivity analysis; affects gating sparsity and performance.
  • Number of experts n = 8
    Set to 8 for all datasets; a hand-chosen hyperparameter, not optimized.
  • dMoE placement (encoder-only) = encoder layers
    Chosen after ablations in Appendix A.5; for transformer it is encoder attention blocks, for CNN after ResBlocks 1-5. This design decision affects results.
assumptions (4)
  • standard math Layer-wise evolution of hidden features approximates Euler discretization of an ODE (Eqs. 8-9).
    Used to set up the optimal control analogy in Section 3.3; standard residual-network approximation.
  • ad hoc to paper The transformation f is linear in its second argument so that the kernel-weighted sum equals a mixture of experts (Eq. 16).
    Invoked to equate kernel smoothing with MoE, but the implemented experts are nonlinear MLPs, so the axiom is false for the actual architecture.
  • domain assumption Attribute flags (race, age band, T-stage) are available at inference and correctly identify subgroup membership.
    Needed for the router to select attribute-specific experts in Eq. (2); deployment may lack such labels.
  • domain assumption Noisy Top-K gating with Gaussian noise yields sufficient exploration and stable training.
    Borrowed from Shazeer et al.; assumed rather than justified for medical segmentation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Distribution-aware Fairness Learning in Medical Image Segmentation From A Control-Theoretic Perspective." pith.science (2026). https://pith.science/paper/YNZX6DOV

@misc{pith2026250200619,
  author       = {Pith},
  title        = {Pith review of: Distribution-aware Fairness Learning in Medical Image Segmentation From A Control-Theoretic Perspective},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YNZX6DOV}},
  note         = {Machine review of arXiv:2502.00619}
}
read the original abstract

Ensuring fairness in medical image segmentation is critical due to biases in imbalanced clinical data acquisition caused by demographic attributes (e.g., age, sex, race) and clinical factors (e.g., disease severity). To address these challenges, we introduce Distribution-aware Mixture of Experts (dMoE), inspired by optimal control theory. We provide a comprehensive analysis of its underlying mechanisms and clarify dMoE's role in adapting to heterogeneous distributions in medical image segmentation. Furthermore, we integrate dMoE into multiple network architectures, demonstrating its broad applicability across diverse medical image analysis tasks. By incorporating demographic and clinical factors, dMoE achieves state-of-the-art performance on two 2D benchmark datasets and a 3D in-house dataset. Our results highlight the effectiveness of dMoE in mitigating biases from imbalanced distributions, offering a promising approach to bridging control theory and medical image segmentation within fairness learning paradigms. The source code will be made available. The source code is available at https://github.com/tvseg/dMoE.

Figures

Figures reproduced from arXiv: 2502.00619 by the authors.

Figure 1
Figure 1. The influence of clinical data distribution on medical image segmentation and the role of dMoE as a distribution-aware control mechanism to address inequity challenges. Transparent blue lines within the violin plots connect the most densely popu￾lated regions of each attribute, visually representing overall equity. tributes such as gender and race, as well as clinical factors, e.g., disease severity, emphasize the n… view at source ↗
Figure 2
Figure 2. (a) Schematic of the dMoE segmentation network for fairness learning, and (b) its interpretation through a control system. 𝐻(𝑥) 𝑖 = (𝑥 ⊤ ⋅𝑊 ) 𝑖+(0, 1)⋅Softplus((𝑥 ⊤ ⋅𝑊 noise) 𝑖 ), (4) KeepTop-𝑘(𝑣, 𝑘) 𝑖 = { 𝑣𝑖 if 𝑣𝑖 is in top 𝑘 elements of 𝑣, −∞ otherwise. (5) where, Softmax(⋅) function normalizes the selected weights, KeepTop-𝑘(⋅) retrains only the top-𝑘 expert contributions, and 𝑊 and 𝑊 noise are trainable weight … view at source ↗
Figure 3
Figure 3. (a) Violin plots depicts attribute-wise segmentation performance. Transparent blue lines within the plots connect the most densely populated regions for each attribute, visually representing overall equity. (b) Qualitative comparison across different subgroups. 8 [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

36 extracted references · 19 canonical work pages

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION format.date year duplicate empty "emp...

  2. [2]

    str \"o m, K. J. Adaptive control. In Mathematical System Theory: The Influence of RE Kalman, pp.\ 437--450. Springer, 1995

  3. [3]

    str \"o m, K. J. and Murray, R. Feedback systems: an introduction for scientists and engineers. Princeton university press, 2021

  4. [4]

    Boskovic, J. D. and Mehra, R. K. Multi-mode switching in flight control. In 19th DASC. 19th Digital Avionics Systems Conference. Proceedings (Cat. No. 00CH37126), volume 2, pp.\ 6F2--1. IEEE, 2000

  5. [5]

    L., and Zhou, Y

    Chen, J., Lu, Y., Yu, Q., Luo, X., Adeli, E., Wang, Y., Lu, L., Yuille, A. L., and Zhou, Y. Transunet: Transformers make strong encoders for medical image segmentation. arXiv preprint arXiv:2102.04306, 2021

  6. [6]

    Low-rank mixture-of-experts for continual medical image segmentation

    Chen, Q., Zhu, L., He, H., Zhang, X., Zeng, S., Ren, Q., and Lu, Y. Low-rank mixture-of-experts for continual medical image segmentation. In International Conference on Medical Image Computing and Computer-Assisted Intervention, pp.\ 382--392. Springer, 2024

  7. [7]

    T., Rubanova, Y., Bettencourt, J., and Duvenaud, D

    Chen, R. T., Rubanova, Y., Bettencourt, J., and Duvenaud, D. K. Neural ordinary differential equations. Advances in neural information processing systems, 31, 2018

  8. [8]

    S., Wicker, J., Sun, Q., and Lee, J

    Chester, A., Koh, Y. S., Wicker, J., Sun, Q., and Lee, J. Balancing utility and fairness against privacy in medical data. In 2020 IEEE Symposium Series on Computational Intelligence (SSCI), pp.\ 1226--1233, 2020. doi:10.1109/SSCI47803.2020.9308226

Show all 36 references
  1. [9]

    S., Brox, T., and Ronneberger, O

    C i c ek, \"O ., Abdulkadir, A., Lienkamp, S. S., Brox, T., and Ronneberger, O. 3D U-Net: learning dense volumetric segmentation from sparse annotation . In Medical Image Computing and Computer-Assisted Intervention, pp.\ 424--432. Springer, 2016

  2. [10]

    C., Francis, B

    Doyle, J. C., Francis, B. A., and Tannenbaum, A. R. Feedback control theory. Courier Corporation, 2013

  3. [11]

    Hofmann, T., Sch \"o lkopf, B., and Smola, A. J. Kernel methods in machine learning. 2008

  4. [12]

    and Shen, Y

    Jiang, Y. and Shen, Y. M4oe: A foundation model for medical multimodal image segmentation with mixture of experts. In International Conference on Medical Image Computing and Computer-Assisted Intervention, pp.\ 621--631. Springer, 2024

  5. [13]

    K., and Li, X

    Jin, R., Xu, Z., Zhong, Y., Yao, Q., Dou, Q., Zhou, S. K., and Li, X. Fairmedfm: fairness benchmarking for medical imaging foundation models. arXiv preprint arXiv:2407.00983, 2024

  6. [14]

    G., Azizi, S., Belgrave, D., Kohli, P., Cemgil, T., et al

    Ktena, I., Wiles, O., Albuquerque, I., Rebuffi, S.-A., Tanno, R., Roy, A. G., Azizi, S., Belgrave, D., Kohli, P., Cemgil, T., et al. Generative models improve fairness of medical classifiers under distribution shifts. Nature Medicine, pp.\ 1--8, 2024

  7. [15]

    Crafting papers on machine learning

    Langley, P. Crafting papers on machine learning. In Langley, P. (ed.), Proceedings of the 17th International Conference on Machine Learning (ICML 2000), pp.\ 1207--1216, Stanford, CA, 2000. Morgan Kaufmann

  8. [16]

    Fairdiff: Fair segmentation with point-image diffusion

    Li, W., Xu, H., Zhang, G., Gao, H.-a., Gao, M., Wang, M., and Zhao, H. Fairdiff: Fair segmentation with point-image diffusion. In International Conference on Medical Image Computing and Computer-Assisted Intervention, pp.\ 617--628. Springer, 2024

  9. [17]

    and Hutter, F

    Loshchilov, I. and Hutter, F. Decoupled weight decay regularization. arXiv preprint arXiv:1711.05101, 2017

  10. [18]

    Beyond finite layer neural networks: Bridging deep architectures and numerical differential equations

    Lu, Y., Zhong, A., Li, Q., and Dong, B. Beyond finite layer neural networks: Bridging deep architectures and numerical differential equations. In International Conference on Machine Learning, pp.\ 3276--3285. PMLR, 2018

  11. [19]

    Learning adversarially fair and transferable representations

    Madras, D., Creager, E., Pitassi, T., and Zemel, R. Learning adversarially fair and transferable representations. In International Conference on Machine Learning, pp.\ 3384--3393. PMLR, 2018

  12. [20]

    K., Cho, Y., Lee, I

    Oh, Y., Park, S., Byun, H. K., Cho, Y., Lee, I. J., Kim, J. S., and Ye, J. C. Llm-driven multimodal target volume contouring in radiation oncology. Nature Communications, 15 0 (1): 0 9186, 2024

  13. [21]

    Pytorch: An imperative style, high-performance deep learning library

    Paszke, A., Gross, S., Massa, F., Lerer, A., Bradbury, J., Chanan, G., Killeen, T., Lin, Z., Gimelshein, N., Antiga, L., et al. Pytorch: An imperative style, high-performance deep learning library. Advances in neural information processing systems, 32, 2019

  14. [22]

    Divide and not forget: Ensemble of selectively trained experts in continual learning

    Rype \'s \'c , G., Cygert, S., Khan, V., Trzci \'n ski, T., Zieli \'n ski, B., and Twardowski, B. Divide and not forget: Ensemble of selectively trained experts in continual learning. arXiv preprint arXiv:2401.10191, 2024

  15. [23]

    W., Hashimoto, T

    Sagawa, S., Koh, P. W., Hashimoto, T. B., and Liang, P. Distributionally robust neural networks for group shifts: On the importance of regularization for worst-case generalization. arXiv preprint arXiv:1911.08731, 2019

  16. [24]

    Outrageously large neural networks: The sparsely-gated mixture-of-experts layer

    Shazeer, N., Mirhoseini, A., Maziarz, K., Davis, A., Le, Q., Hinton, G., and Dean, J. Outrageously large neural networks: The sparsely-gated mixture-of-experts layer. arXiv preprint arXiv:1701.06538, 2017

  17. [25]

    Layer-parallel training of residual networks with auxiliary variable networks

    Sun, Q., Dong, H., Chen, Z., Sun, J., Li, Z., and Dong, B. Layer-parallel training of residual networks with auxiliary variable networks. Numerical Methods for Partial Differential Equations, 40 0 (6): 0 e23147, 2024

  18. [26]

    A., Branche, A

    Theodore, D. A., Branche, A. R., Zhang, L., Graciaa, D. S., Choudhary, M., Hatlen, T. J., Osman, R., Babu, T. M., Robinson, S. T., Gilbert, P. B., et al. Clinical and demographic factors associated with covid-19, severe covid-19, and sars-cov-2 infection in adults: a secondary...

  19. [27]

    Fairseg: A large-scale medical image segmentation dataset for fairness learning using segment anything model with fair error-bound scaling

    Tian, Y., Shi, M., Luo, Y., Kouhana, A., Elze, T., and Wang, M. Fairseg: A large-scale medical image segmentation dataset for fairness learning using segment anything model with fair error-bound scaling. In The Twelfth International Conference on Learning Representations, 2024

  20. [28]

    M., Huang, H., Khan, M

    Tian, Y., Wen, C., Shi, M., Afzal, M. M., Huang, H., Khan, M. O., Luo, Y., Fang, Y., and Wang, M. Fairdomain: Achieving fairness in cross-domain medical image segmentation and classification. In European Conference on Computer Vision, pp.\ 251--271. Springer, 2025

  21. [29]

    The ham10000 dataset, a large collection of multi-source dermatoscopic images of common pigmented skin lesions

    Tschandl, P., Rosendahl, C., and Kittler, H. The ham10000 dataset, a large collection of multi-source dermatoscopic images of common pigmented skin lesions. Scientific data, 5 0 (1): 0 1--9, 2018

  22. [30]

    N., Kaiser, ., and Polosukhin, I

    Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, ., and Polosukhin, I. Attention is all you need. Advances in neural information processing systems, 30, 2017

  23. [31]

    Sam-med3d-moe: Towards a non-forgetting segment anything model via mixture of experts for 3d medical image segmentation

    Wang, G., Ye, J., Cheng, J., Li, T., Chen, Z., Cai, J., He, J., and Zhuang, B. Sam-med3d-moe: Towards a non-forgetting segment anything model via mixture of experts for 3d medical image segmentation. In International Conference on Medical Image Computing and Computer-Assisted ...

  24. [32]

    A proposal on machine learning via dynamical systems

    Weinan, E. A proposal on machine learning via dynamical systems. Communications in Mathematics and Statistics, 1 0 (5): 0 1--11, 2017

  25. [33]

    Mode switching control design with initial value compensation and its application to head positioning control on magnetic disk drives

    Yamaguchi, T., Shishida, K., Tohyama, S., and Hirai, H. Mode switching control design with initial value compensation and its application to head positioning control on magnetic disk drives. IEEE Transactions on Industrial Electronics, 43 0 (1): 0 65--73, 1996

  26. [34]

    Boosting continual learning of vision-language models via mixture-of-experts adapters

    Yu, J., Zhuge, Y., Zhang, L., Hu, P., Wang, D., Lu, H., and He, Y. Boosting continual learning of vision-language models via mixture-of-experts adapters. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp.\ 23219--23230, 2024

  27. [35]

    Sliding mode switching control of manipulators based on disturbance observer

    Yu, L., Huang, J., and Fei, S. Sliding mode switching control of manipulators based on disturbance observer. Circuits, Systems, and Signal Processing, 36: 0 2574--2585, 2017

  28. [36]

    D., Visentin, M., Qiao, M., Gu, R., Ouyang, C., Liu, Y., Matthews, P

    Zhang, X., Ou, N., Basaran, B. D., Visentin, M., Qiao, M., Gu, R., Ouyang, C., Liu, Y., Matthews, P. M., Ye, C., et al. A foundation model for brain lesion segmentation with mixture of modality experts. In International Conference on Medical Image Computing and Computer-Assist...

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.