pith. sign in

arxiv: 2606.08687 · v1 · pith:3L5XK6WWnew · submitted 2026-06-07 · 💻 cs.CV

Shift-Dependent Asymmetry: Orthogonal Inverse Low-Rank Adaptation for Federated Medical Segmentation

Xingyue Zhao , Wenke Huang , Linghao Zhuang , Haoran Wu , Anwen Jiang , Zhifeng Wang , Wenwen He , Ming Feng
show 2 more authors
Mang Ye Bo Xu
This is my paper

Pith reviewed 2026-06-27 18:25 UTC · model grok-4.3

classification 💻 cs.CV
keywords federated learningLoRAmedical image segmentationasymmetric adaptationdomain shiftparameter-efficient fine-tuningorthogonality regularization
0
0 comments X

The pith

Encoder-decoder asymmetry in medical segmentation requires separate personalization paths in federated LoRA to keep site biases from leaking into shared anatomy.

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

The paper shows that uniform aggregation rules in federated LoRA break down for medical segmentation because the encoder mainly sees appearance shifts across sites while the decoder mainly sees differences in supervision. IAT counters this by routing personalization to module-specific components in the encoder for shifts and in the decoder for supervision, while preserving a shared pathway that carries consensus anatomy. A bilinear coupling in LoRA can still let site-specific updates leak into the shared direction, so the method adds a Subspace Orthogonality Regularizer that penalizes collinearity between shared and local update subspaces. The approach improves generalization across institutions without raising communication costs. A reader would care because medical imaging data stays siloed by hospital, and any method that respects this asymmetry can deliver usable models from distributed training.

Core claim

IAT aligns adaptation with heterogeneity sources by personalizing module-specific components in the encoder to absorb appearance shifts and in the decoder to accommodate site-dependent supervision, while retaining a shared pathway for transferable consensus. Under LoRA's bilinear parameterization, multiplicative coupling can still cause site-specific updates to leak into the shared direction, so a Subspace Orthogonality Regularizer penalizes shared-local collinearity in the effective update space, mitigating leakage without extra communication.

What carries the argument

Inverse Asymmetric Tuning (IAT) with module-specific personalization in encoder and decoder plus a Subspace Orthogonality Regularizer that enforces orthogonality between shared and local update subspaces.

If this is right

  • Personalization confined to encoder modules absorbs appearance shifts while the shared pathway preserves consensus anatomy.
  • Personalization confined to decoder modules accommodates site-dependent supervision without contaminating the shared model.
  • The orthogonality regularizer prevents site-specific leakage from bilinear coupling without requiring additional communication rounds.
  • The resulting models show consistent gains over uniform federated LoRA and other parameter-efficient baselines on medical segmentation tasks.

Where Pith is reading between the lines

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

  • The same asymmetry logic could be tested on other encoder-decoder tasks such as detection or reconstruction where heterogeneity sources also split along the network depth.
  • If the regularizer proves robust, it might serve as a drop-in addition to any LoRA-based federated method that already separates parameters by module type.
  • The approach leaves open whether explicit modeling of shift sources at every layer would yield further gains or whether the current coarse encoder-decoder split is already near-optimal.

Load-bearing premise

Encoder-decoder asymmetry is the dominant source of entanglement and penalizing collinearity in the LoRA update space is enough to block site-specific leakage into the shared direction without harming convergence.

What would settle it

Run the same federated segmentation tasks with and without the orthogonality regularizer on datasets that exhibit documented encoder appearance shifts and decoder supervision differences; if performance and cross-site generalization remain statistically identical, the leakage-mitigation claim does not hold.

Figures

Figures reproduced from arXiv: 2606.08687 by Anwen Jiang, Bo Xu, Haoran Wu, Linghao Zhuang, Mang Ye, Ming Feng, Wenke Huang, Wenwen He, Xingyue Zhao, Zhifeng Wang.

Figure 1
Figure 1. Figure 1: Illustration of the motivation. (a) Standard FedLoRA employs naive aggregation of all parameters, suffering from high communication costs and severe client drift. (b) Existing Methods (e.g., uniform partial sharing) enforce a static splitting rule across the entire network, overlooking the distinct roles of different mod￾ules. (c) Medical Challenge: Encoder-Decoder Asymmetry. We identify a critical structu… view at source ↗
Figure 2
Figure 2. Figure 2: Architecture illustration of the proposed framework. To address the structural misalignment between covariate and concept shifts, we implement Inverse Asymmetric Tuning (IAT) which allocates trainable LoRA factors inversely across the encoder and decoder. To prevent the coupling of shared knowledge with local biases, we apply a Subspace Orthogonality Regularizer (SOR) during local updates. The server aggre… view at source ↗
Figure 3
Figure 3. Figure 3: Preliminary ablation on LoRA configuration. We compare the performance of applying LoRA to the Encoder-only versus the Encoder+Decoder. The results demonstrate that, unlike classification tasks, medical segmentation requires adapting the decoder to reconstruct precise pixel-level details, validating our choice to inject LoRA into both modules. Assumption 3.3 (Non-Degenerate LoRA Factors). To en￾sure suffic… view at source ↗
Figure 4
Figure 4. Figure 4: Sensitivity analysis of λ. Left: Histology nuclei. Right: Fundus photography. 3.59% on TNBC), while FedIT fails to generalize to distinct domains like Drishti-GS1 (57.74%). In contrast, our method successfully maintains robust generalization capabilities un￾der these challenging conditions. By effectively decoupling covariate and concept shifts, our framework achieves the highest average DSC of 81.40% and … view at source ↗
read the original abstract

Low-Rank Adaptation (LoRA) enables efficient federated fine-tuning of segmentation foundation models for medical imaging. However, most federated LoRA methods adopt a uniform aggregation rule, which breaks under the encoder-decoder asymmetry in medical segmentation: the encoder is dominated by appearance shifts, while the decoder is dominated by supervision variations. This mismatch entangles shared anatomy with site-specific biases and harms generalization. To address this, we propose Inverse Asymmetric Tuning (IAT). IAT aligns adaptation with heterogeneity sources by personalizing module-specific components in the encoder to absorb appearance shifts and in the decoder to accommodate site-dependent supervision, while retaining a shared pathway for transferable consensus. However, structural separation alone is insufficient under LoRA's bilinear parameterization, where multiplicative coupling can still cause site-specific updates to leak into the shared direction. We therefore introduce a Subspace Orthogonality Regularizer that penalizes shared-local collinearity in the effective update space, mitigating leakage without extra communication. Experiments show consistent improvements over strong federated LoRA and parameter-efficient FL baselines.

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 proposes Inverse Asymmetric Tuning (IAT) for federated LoRA fine-tuning of medical segmentation models. It identifies encoder-decoder asymmetry arising from appearance shifts (encoder) versus supervision variations (decoder), addresses it by personalizing module-specific components while keeping a shared pathway, and introduces a Subspace Orthogonality Regularizer to penalize shared-local collinearity in the effective LoRA update space and thereby reduce site-specific leakage without extra communication. The abstract states that experiments demonstrate consistent improvements over federated LoRA and parameter-efficient FL baselines.

Significance. If the claimed improvements hold under rigorous evaluation, the method would address a practical challenge in federated medical imaging by aligning adaptation structure with known sources of heterogeneity. The absence of extra communication is a potential strength for real-world deployment, though this remains unverified from the given description.

major comments (2)
  1. [Abstract] Abstract: the central claim that 'experiments show consistent improvements' supplies no quantitative results, ablation studies, dataset descriptions, statistical tests, or baseline numbers, rendering the effectiveness of IAT and the regularizer impossible to assess against evidence.
  2. [Abstract] Abstract: no derivation or analysis is supplied showing that the Subspace Orthogonality Regularizer is strong enough to counteract multiplicative coupling in LoRA updates and block leakage into the shared direction without harming convergence across heterogeneous sites.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address each major comment point by point below, drawing on the content of the full manuscript.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim that 'experiments show consistent improvements' supplies no quantitative results, ablation studies, dataset descriptions, statistical tests, or baseline numbers, rendering the effectiveness of IAT and the regularizer impossible to assess against evidence.

    Authors: Abstracts are conventionally high-level summaries that state claims without embedding full quantitative details, ablations, or statistics to preserve brevity and readability. The full manuscript supplies all requested elements in Sections 4 and 5: quantitative Dice and HD improvements over federated LoRA and PEFT baselines, ablation studies isolating the contribution of module-specific personalization and the orthogonality regularizer, descriptions of the medical segmentation datasets used, and statistical significance testing. We can incorporate one or two key numerical highlights into the abstract during revision if the editor prefers. revision: partial

  2. Referee: [Abstract] Abstract: no derivation or analysis is supplied showing that the Subspace Orthogonality Regularizer is strong enough to counteract multiplicative coupling in LoRA updates and block leakage into the shared direction without harming convergence across heterogeneous sites.

    Authors: Section 3.3 derives the Subspace Orthogonality Regularizer explicitly as a penalty on the cosine similarity between the effective shared and local LoRA update directions, showing how this term directly counters the bilinear (multiplicative) coupling that would otherwise allow site-specific components to leak into the aggregated pathway. The accompanying analysis explains why the regularizer preserves convergence by operating only on the update subspace without altering the base model or requiring additional communication. While we do not supply a general convergence theorem for arbitrary heterogeneity levels, the empirical results across multiple sites confirm stable training; we are prepared to add further analytic bounds in an appendix if requested. revision: no

Circularity Check

0 steps flagged

No equations or derivations present; derivation chain cannot be inspected for reduction

full rationale

The provided abstract and manuscript excerpt contain only conceptual descriptions of IAT and the Subspace Orthogonality Regularizer with no equations, parameter fittings, or derivation steps shown. Without any mathematical content that could reduce a claimed prediction to an input by construction, self-citation, or ansatz, no circular steps exist to flag. The paper's claims remain at the level of architectural motivation and empirical improvement statements, which are self-contained against external benchmarks in the absence of internal reductions.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 2 invented entities

Abstract-only; no explicit free parameters, axioms, or invented entities are quantified. The core modeling choice (encoder vs decoder dominance) is treated as a domain assumption.

axioms (1)
  • domain assumption Encoder is dominated by appearance shifts while decoder is dominated by supervision variations
    Stated directly in the abstract as the source of the mismatch that uniform aggregation breaks.
invented entities (2)
  • Inverse Asymmetric Tuning (IAT) no independent evidence
    purpose: Personalize module-specific components separately in encoder and decoder while retaining shared pathway
    New method introduced to align adaptation with heterogeneity sources
  • Subspace Orthogonality Regularizer no independent evidence
    purpose: Penalize shared-local collinearity to mitigate leakage in LoRA bilinear updates
    Introduced to address multiplicative coupling without extra communication

pith-pipeline@v0.9.1-grok · 5740 in / 1341 out tokens · 28016 ms · 2026-06-27T18:25:46.950340+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

114 extracted references · 1 canonical work pages

  1. [1]

    Communication, Simulation, and Intelligent Agents: Implications of Personal Intelligent Machines for Medical Education

    Clancey, William J. Communication, Simulation, and Intelligent Agents: Implications of Personal Intelligent Machines for Medical Education. Proceedings of the Eighth International Joint Conference on Artificial Intelligence (IJCAI-83)

  2. [2]

    Classification Problem Solving

    Clancey, William J. Classification Problem Solving. Proceedings of the Fourth National Conference on Artificial Intelligence

  3. [3]

    , title =

    Robinson, Arthur L. , title =. 1980 , doi =. https://science.sciencemag.org/content/208/4447/1019.full.pdf , journal =

    1980

  4. [4]

    New Ways to Make Microcircuits Smaller---Duplicate Entry

    Robinson, Arthur L. New Ways to Make Microcircuits Smaller---Duplicate Entry. Science

  5. [5]

    Clancey and Glenn Rennels , abstract =

    Diane Warner Hasling and William J. Clancey and Glenn Rennels , abstract =. Strategic explanations for a diagnostic consultation system , journal =. 1984 , issn =. doi:https://doi.org/10.1016/S0020-7373(84)80003-6 , url =

    work page doi:10.1016/s0020-7373(84)80003-6 1984

  6. [6]

    and Rennels, Glenn R

    Hasling, Diane Warner and Clancey, William J. and Rennels, Glenn R. and Test, Thomas. Strategic Explanations in Consultation---Duplicate. The International Journal of Man-Machine Studies

  7. [7]

    Poligon: A System for Parallel Problem Solving

    Rice, James. Poligon: A System for Parallel Problem Solving

  8. [8]

    Transfer of Rule-Based Expertise through a Tutorial Dialogue

    Clancey, William J. Transfer of Rule-Based Expertise through a Tutorial Dialogue

  9. [9]

    The Engineering of Qualitative Models

    Clancey, William J. The Engineering of Qualitative Models

  10. [10]

    Advances in neural information processing systems , volume=

    Attention is all you need , author=. Advances in neural information processing systems , volume=

  11. [11]

    Pluto: The 'Other' Red Planet

    NASA. Pluto: The 'Other' Red Planet

  12. [12]

    AISTATS , pages=

    Communication-Efficient Learning of Deep Networks from Decentralized Data , author=. AISTATS , pages=

  13. [13]

    Proceedings of Machine learning and systems , volume=

    Federated optimization in heterogeneous networks , author=. Proceedings of Machine learning and systems , volume=

  14. [14]

    CVPR , pages=

    Model-Contrastive Federated Learning , author=. CVPR , pages=

  15. [15]

    AAAI , year=

    FedProto: Federated Prototype Learning across Heterogeneous Clients , author=. AAAI , year=

  16. [16]

    CVPR , month =

    Huang, Wenke and Ye, Mang and Shi, Zekun and Li, He and Du, Bo , title =. CVPR , month =. 2023 , pages =

    2023

  17. [17]

    CVPR , month =

    Son, Ha Min and Kim, Moon-Hyun and Chung, Tai-Myoung and Huang, Chao and Liu, Xin , title =. CVPR , month =. 2024 , pages =

    2024

  18. [18]

    NeurIPS , year=

    Taming Cross-Domain Representation Variance in Federated Prototype Learning with Heterogeneous Data Domains , author=. NeurIPS , year=

  19. [19]

    FedContrast-GPA: Heterogeneous Federated Optimization via Local Contrastive Learning and Global Process-Aware Aggregation

    Zhou, Qin and Zheng, Guoyan. FedContrast-GPA: Heterogeneous Federated Optimization via Local Contrastive Learning and Global Process-Aware Aggregation. MICCAI. 2023

    2023

  20. [20]

    Jianqiao Zhang and Caifeng Shan and Jungong Han , booktitle=. Fed. 2024 , url=

    2024

  21. [21]

    IEEE TCSVT , year=

    Ultrasound Nodule Segmentation Using Asymmetric Learning with Simple Clinical Annotation , author=. IEEE TCSVT , year=

  22. [22]

    IEEE TMI , volume=

    HAMIL: High-resolution activation maps and interleaved learning for weakly supervised segmentation of histopathological images , author=. IEEE TMI , volume=. 2023 , publisher=

    2023

  23. [23]

    CVPR , pages=

    Learn from others and be yourself in heterogeneous federated learning , author=. CVPR , pages=

  24. [24]

    CVPR , pages=

    Fair Federated Learning under Domain Skew with Local Consistency and Domain Diversity , author=. CVPR , pages=

  25. [25]

    CVPR , pages=

    Fair federated medical image segmentation via client contribution estimation , author=. CVPR , pages=

  26. [26]

    FedBN: Federated Learning on Non-IID Features via Local Batch Normalization , author=

  27. [27]

    IEEE TPAMI , year=

    Generalizable heterogeneous federated cross-correlation and instance similarity learning , author=. IEEE TPAMI , year=

  28. [28]

    CVPR , pages=

    FedDG: Federated Domain Generalization on Medical Image Segmentation via Episodic Learning in Continuous Frequency Space , author=. CVPR , pages=

  29. [29]

    CVPR , pages=

    Federated domain generalization with generalization adjustment , author=. CVPR , pages=

  30. [30]

    International conference on machine learning , pages=

    Ditto: Fair and robust federated learning through personalization , author=. International conference on machine learning , pages=. 2021 , organization=

    2021

  31. [31]

    AAAI , volume=

    Fairfed: Enabling group fairness in federated learning , author=. AAAI , volume=

  32. [32]

    AAAI , volume=

    Harmofl: Harmonizing local and global drifts in federated learning on heterogeneous medical images , author=. AAAI , volume=

  33. [33]

    Medical Image Analysis , volume=

    Semi-supervised medical image segmentation via uncertainty rectified pyramid consistency , author=. Medical Image Analysis , volume=. 2022 , publisher=

    2022

  34. [34]

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

    Contextrast: Contextual Contrastive Learning for Semantic Segmentation , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  35. [35]

    CVPR , pages=

    Decompose, adjust, compose: Effective normalization by playing with frequency for domain generalization , author=. CVPR , pages=

  36. [36]

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

    Amplitude-phase recombination: Rethinking robustness of convolutional neural networks in frequency domain , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  37. [37]

    AAAI , volume=

    Federated graph learning under domain shift with generalizable prototypes , author=. AAAI , volume=

  38. [38]

    CVPR , pages=

    Think Twice Before Selection: Federated Evidential Active Learning for Medical Image Analysis with Domain Shifts , author=. CVPR , pages=

  39. [39]

    Nature Methods , volume=

    D-LMBmap: a fully automated deep-learning pipeline for whole-brain profiling of neural circuitry , author=. Nature Methods , volume=. 2023 , publisher=

    2023

  40. [40]

    IEEE TMI , volume=

    IOP-FL: Inside-outside personalization for federated medical image segmentation , author=. IEEE TMI , volume=. 2023 , publisher=

    2023

  41. [41]

    arXiv preprint arXiv:1610.02527 , year=

    Federated optimization: Distributed machine learning for on-device intelligence , author=. arXiv preprint arXiv:1610.02527 , year=

    Pith/arXiv arXiv

  42. [42]

    ECCV , pages=

    Personalizing federated medical image segmentation via local calibration , author=. ECCV , pages=. 2022 , organization=

    2022

  43. [43]

    CVPR , pages=

    Rethinking the Representation in Federated Unsupervised Learning with Non-IID Data , author=. CVPR , pages=

  44. [44]

    MICCAI , pages=

    Black-box Domain Adaptative Cell Segmentation via Multi-source Distillation , author=. MICCAI , pages=. 2023 , organization=

    2023

  45. [45]

    IEEE transactions on medical imaging , volume=

    MS-Net: multi-site network for improving prostate segmentation with heterogeneous MRI data , author=. IEEE transactions on medical imaging , volume=. 2020 , publisher=

    2020

  46. [46]

    Computers in biology and medicine , volume=

    Computer-aided detection and diagnosis for prostate cancer based on mono and multi-parametric MRI: a review , author=. Computers in biology and medicine , volume=. 2015 , publisher=

    2015

  47. [47]

    Medical image analysis , volume=

    Evaluation of prostate segmentation algorithms for MRI: the PROMISE12 challenge , author=. Medical image analysis , volume=. 2014 , publisher=

    2014

  48. [48]

    ieee-isbi conf

    Nci-proc. ieee-isbi conf. 2013 challenge: Automated segmentation of prostate structures , author=. 2015 , howpublished=

    2013

  49. [49]

    Pacific symposium on biocomputing Co-chairs , pages=

    Crowdsourcing image annotation for nucleus detection and segmentation in computational pathology: evaluating experts, automated methods, and the crowd , author=. Pacific symposium on biocomputing Co-chairs , pages=. 2014 , organization=

    2014

  50. [50]

    Scientific reports , volume=

    Glandular morphometrics for objective grading of colorectal adenocarcinoma histology images , author=. Scientific reports , volume=. 2017 , publisher=

    2017

  51. [51]

    Medical image computing and computer-assisted intervention--MICCAI 2015: 18th international conference, Munich, Germany, October 5-9, 2015, proceedings, part III 18 , pages=

    U-net: Convolutional networks for biomedical image segmentation , author=. Medical image computing and computer-assisted intervention--MICCAI 2015: 18th international conference, Munich, Germany, October 5-9, 2015, proceedings, part III 18 , pages=. 2015 , organization=

    2015

  52. [52]

    Scientific data , volume=

    Dataset of segmented nuclei in hematoxylin and eosin stained histopathology images of ten cancer types , author=. Scientific data , volume=. 2020 , publisher=

    2020

  53. [53]

    CVPR , pages=

    Efficient parameter-free clustering using first neighbor relations , author=. CVPR , pages=

  54. [54]

    International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=

    Morestyle: relax low-frequency constraint of fourier-based image reconstruction in generalizable medical image segmentation , author=. International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=. 2024 , organization=

    2024

  55. [55]

    CVPR , year=

    Geometric Knowledge-Guided Localized Global Distribution Alignment for Federated Learning , author=. CVPR , year=

  56. [56]

    TPAMI , year=

    A Federated Learning for Generalization, Robustness, Fairness: A Survey and Benchmark , author=. TPAMI , year=

  57. [57]

    2024 IEEE International Symposium on Biomedical Imaging (ISBI) , pages=

    Sam-Driven Weakly Supervised Nodule Segmentation with Uncertainty-Aware Cross Teaching , author=. 2024 IEEE International Symposium on Biomedical Imaging (ISBI) , pages=. 2024 , organization=

    2024

  58. [58]

    Neural Networks , volume=

    Expert guidance and partially-labeled data collaboration for multi-organ segmentation , author=. Neural Networks , volume=. 2025 , publisher=

    2025

  59. [59]

    ACM Computing Surveys , volume=

    Heterogeneous federated learning: State-of-the-art and research challenges , author=. ACM Computing Surveys , volume=. 2023 , publisher=

    2023

  60. [60]

    Meta-Radiology , volume=

    Cardiovascular medical image and analysis based on 3d vision: A comprehensive survey , author=. Meta-Radiology , volume=. 2024 , publisher=

    2024

  61. [61]

    International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=

    WIA-LD2ND: Wavelet-based image alignment for self-supervised low-dose CT denoising , author=. International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=. 2024 , organization=

    2024

  62. [62]

    arXiv preprint arXiv:2502.08200 , year=

    ActiveSSF: An Active-Learning-Guided Self-Supervised Framework for Long-Tailed Megakaryocyte Classification , author=. arXiv preprint arXiv:2502.08200 , year=

    arXiv

  63. [63]

    Proceedings of the IEEE/CVF international conference on computer vision , pages=

    Segment anything , author=. Proceedings of the IEEE/CVF international conference on computer vision , pages=

  64. [64]

    , author=

    Lora: Low-rank adaptation of large language models. , author=. ICLR , volume=

  65. [65]

    International Conference on Machine Learning , pages=

    The non-iid data quagmire of decentralized machine learning , author=. International Conference on Machine Learning , pages=. 2020 , organization=

    2020

  66. [66]

    Foundations and trends

    Advances and open problems in federated learning , author=. Foundations and trends. 2021 , publisher=

    2021

  67. [67]

    Advances in neural information processing systems , volume=

    Tackling the objective inconsistency problem in heterogeneous federated optimization , author=. Advances in neural information processing systems , volume=

  68. [68]

    International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=

    Fedfms: Exploring federated foundation models for medical image segmentation , author=. International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=. 2024 , organization=

    2024

  69. [69]

    International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=

    A federated learning-friendly approach for parameter-efficient fine-tuning of sam in 3d segmentation , author=. International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=. 2024 , organization=

    2024

  70. [70]

    ICASSP 2026-2026 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages=

    pFedSAM: Personalized federated learning of Segment Anything Model for medical image segmentation , author=. ICASSP 2026-2026 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages=. 2026 , organization=

    2026

  71. [71]

    IEEE Transactions on Information Forensics and Security , year=

    Federated Client-tailored Adapter for Medical Image Segmentation , author=. IEEE Transactions on Information Forensics and Security , year=

  72. [72]

    International conference on machine learning , pages=

    Exploiting shared representations for personalized federated learning , author=. International conference on machine learning , pages=. 2021 , organization=

    2021

  73. [73]

    9th International Conference on Learning Representations (

    Xiaoxiao Li and Meirui Jiang and Xiaofei Zhang and Michael Kamp and Qi Dou , title =. 9th International Conference on Learning Representations (

  74. [74]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Fedas: Bridging inconsistency in personalized federated learning , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  75. [75]

    arXiv preprint arXiv:2310.10049 , year=

    Fate-llm: A industrial grade federated learning framework for large language models , author=. arXiv preprint arXiv:2310.10049 , year=

    arXiv

  76. [76]

    2024 USENIX Annual Technical Conference (USENIX ATC 24) , pages=

    \ FwdLLM \ : Efficient federated finetuning of large language models with perturbed inferences , author=. 2024 USENIX Annual Technical Conference (USENIX ATC 24) , pages=

    2024

  77. [77]

    European Conference on Computer Vision , pages=

    Learn to preserve and diversify: Parameter-efficient group with orthogonal regularization for domain generalization , author=. European Conference on Computer Vision , pages=. 2024 , organization=

    2024

  78. [78]

    International conference on machine learning , pages=

    Parameter-efficient transfer learning for NLP , author=. International conference on machine learning , pages=. 2019 , organization=

    2019

  79. [79]

    Advances in Neural Information Processing Systems , volume=

    Few-shot parameter-efficient fine-tuning is better and cheaper than in-context learning , author=. Advances in Neural Information Processing Systems , volume=

  80. [80]

    arXiv preprint arXiv:2310.13283 , year=

    pFedLoRA: Model-heterogeneous personalized federated learning with LoRA tuning , author=. arXiv preprint arXiv:2310.13283 , year=

    arXiv

Showing first 80 references.