pith. sign in

arxiv: 2605.17508 · v1 · pith:G36RVS7Snew · submitted 2026-05-17 · 💻 cs.LG · cs.AI

BESplit: Bias-Compensated Split Federated Learning with Evidential Aggregation

Yuhan Xie , Chen Lyu , Jingrong Huang This is my paper

Pith reviewed 2026-05-20 14:23 UTC · model grok-4.3

classification 💻 cs.LG cs.AI
keywords split federated learningnon-IID databias compensationevidential aggregationmodel partitioningcollaborative trainingprivacy-preserving learning
0
0 comments X

The pith

Split federated learning can reduce non-IID bias by reweighting client updates with evidential uncertainty and pairing complementary clients.

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

The paper sets out to establish that the split structure in federated learning creates new ways to handle biased optimization when client data distributions differ. It builds a framework that reweights each client's contribution according to uncertainty estimates, aligns representations by matching clients with offsetting data patterns, and synchronizes the separated model parts so they can still train coherently. If these steps hold, collaborative models would converge more reliably and reach higher accuracy without clients exposing their raw data. A reader would care because non-IID data is the norm in device networks, and existing fixes borrowed from standard federated learning often leave residual skew and instability.

Core claim

BESplit exploits the split architecture of SFL to perform bias compensation beyond ordinary parameter averaging: Evidential Aggregation reweights client contributions using fine-grained uncertainty measures to stop biased local data from dominating global updates, Bias-Compensated Collaboration pairs clients with complementary distributions to align split-layer representations and reduce distributional skew, and Dual-Teacher Distillation keeps the decoupled client and server models synchronized so each can perform independent local inference.

What carries the argument

Evidential Aggregation combined with Bias-Compensated Collaboration, which together reweight contributions by uncertainty and align representations through complementary client pairing inside the split model structure.

If this is right

  • Biased local data stops dominating global updates because client contributions receive fine-grained reweighting based on evidential uncertainty.
  • Distributional skew decreases when split-layer representations are aligned by deliberately pairing clients whose data patterns offset each other.
  • Client and server models can run independent local inference once dual-teacher distillation has synchronized their knowledge.
  • Training reaches higher accuracy with more stable convergence and lower computational cost across varied non-IID partitions.

Where Pith is reading between the lines

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

  • The client-pairing idea could be adapted to other partitioned training setups where model sections are held on separate devices.
  • Evidential uncertainty estimates might also serve as a lightweight signal for identifying low-quality clients without extra privacy leakage.
  • If the pairing mechanism scales, it could lower the amount of manual tuning needed when deploying split learning on new device fleets.

Load-bearing premise

The split architecture of SFL inherently changes how client information is represented and coordinated in ways that allow bias compensation beyond standard parameter-level aggregation.

What would settle it

An ablation experiment on the same five benchmark datasets that replaces evidential reweighting and complementary pairing with ordinary averaging and random pairing, then measures whether accuracy and convergence gains vanish under non-IID conditions.

Figures

Figures reproduced from arXiv: 2605.17508 by Chen Lyu, Jingrong Huang, Yuhan Xie.

Figure 1
Figure 1. Figure 1: Illustration of BESplit. EA, BCC, and DTD modules are integrated into the SFL workflow. In each communication round, client Ck processes a mini-batch of local samples through WCk to produce smashed data, which are sent to the corresponding WSk for evidential prediction. The server records these outputs in a client-specific CSR, capturing fine-grained prediction statistics. Based on CSR updates, EA adaptive… view at source ↗
Figure 2
Figure 2. Figure 2: SFL process pipeline with parallel knowledge distillation. Note: The color scheme of the components is consistent with that used in [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: BMR comparison under varying data heterogeneity on HAM10000. optimization signal that approximates training under a more balanced feature distribution. It is worth noting that BCC introduces no negative effects even if a suitable complementary client is momentarily un￾available under extreme conditions. In such cases, the sys￾tem does not wait; instead, BCC is simply skipped for that batch, seamlessly fall… view at source ↗
Figure 4
Figure 4. Figure 4: Accuracy comparison of BESplit-G and BESplit-A under different non-IID levels. 5. Experiments 5.1. Experimental Setup Data. We evaluate BESplit on five widely used datasets, including two medical imaging datasets, ISIC (Ogier du Terrail et al., 2022) and HAM10000 (Tschandl et al., 2018), and three natural image datasets, F-MNIST (Xiao et al., 2017), CIFAR10, and CIFAR100 (Krizhevsky et al., 2009). Detailed… view at source ↗
Figure 5
Figure 5. Figure 5: Comparison of TTA on the HAM10000 and CIFAR10 datasets. maintains strong performance, achieving 1.3% higher accu￾racy than the best baseline under the default non-IID setting and performing closely to BESplit-G, demonstrating its ef￾fectiveness and deployability in practical SFL systems. 5.3. Ablation Study We conduct ablation studies on five datasets to assess the contribution of each component in BESplit… view at source ↗
Figure 6
Figure 6. Figure 6: Sensitivity analysis of BESplit on the F-MNIST dataset. tions, and the gap widens as the accuracy target increases, reflecting accumulated optimization bias. In contrast, BE￾Split maintains stable improvement throughout training and achieves the lowest RTA across all targets, with particularly pronounced gains under non-IID scenarios. This demon￾strates that BESplit achieves fast and stable convergence, ef… view at source ↗
Figure 7
Figure 7. Figure 7: BMR comparison under varying degrees of data heterogeneity on medical datasets (HAM10000 and ISIC). F.3. Robustness Against Extreme Data Heterogeneity The training behaviors of BESplit-G and BESplit-A under increasing levels of non-IID data heterogeneity are shown in [PITH_FULL_IMAGE:figures/full_fig_p020_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Training curves of BESplit-G and BESplit-A under varying non-IID level on the HAM10000 dataset. F.4. Model Size Evaluation As shown in [PITH_FULL_IMAGE:figures/full_fig_p021_8.png] view at source ↗
read the original abstract

Split Federated Learning (SFL) enables privacy-preserving collaborative training by partitioning models between clients and a server. However, under non-IID data distributions, SFL often suffers from biased optimization and unstable convergence, while existing solutions largely adapt techniques from conventional federated learning. In this work, we observe that the split architecture of SFL inherently alters how client information is represented and coordinated, opening opportunities for bias compensation beyond parameter-level aggregation. Based on this insight, we propose BESplit, an architecture-aware framework that exploits the intrinsic structure of SFL to mitigate non-IID effects. First, to prevent biased local data from dominating global updates, we introduce Evidential Aggregation (EA) to perform fine-grained reweighting of client contributions based on evidential uncertainty. Second, to further reduce distributional skew, we develop Bias-Compensated Collaboration (BCC) to align split-layer representations by pairing complementary clients. Finally, Dual-Teacher Distillation (DTD) is incorporated to synchronize knowledge between decoupled client and server models, enabling independent local inference. Extensive experiments on five benchmark datasets demonstrate that BESplit consistently outperforms state-of-the-art methods in accuracy, convergence stability, and computational efficiency under diverse non-IID settings.

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 / 2 minor

Summary. The paper proposes BESplit, a framework for Split Federated Learning (SFL) under non-IID data. It introduces three components motivated by the claim that SFL's split architecture inherently changes client information representation and coordination: Evidential Aggregation (EA) for uncertainty-based reweighting of client updates, Bias-Compensated Collaboration (BCC) for pairing complementary clients to align split-layer representations, and Dual-Teacher Distillation (DTD) for synchronizing knowledge between decoupled client and server models to enable independent inference. The central empirical claim is that BESplit consistently outperforms prior SFL methods in accuracy, convergence stability, and efficiency across five benchmark datasets under diverse non-IID partitions.

Significance. If the results hold after addressing the ablation gap, the work could meaningfully advance SFL by demonstrating how split-model structure enables bias compensation mechanisms beyond standard parameter aggregation. The use of evidential uncertainty for client reweighting and the distillation approach for decoupled models represent potentially useful technical ideas for handling heterogeneity in privacy-preserving settings.

major comments (2)
  1. [Experiments] Experiments section: The comparisons are limited to existing SFL adaptations. No ablation applies the identical EA/BCC/DTD components inside a non-split baseline such as FedAvg or FedProx (with all other hyperparameters fixed). This leaves open the possibility that reported gains arise from the reweighting and distillation mechanisms themselves rather than from any split-specific representational change, directly undermining the architecture-aware premise stated in the abstract and introduction.
  2. [Introduction and §3] Introduction and §3: The observation that 'the split architecture of SFL inherently alters how client information is represented and coordinated' is presented as the foundation for the three techniques, yet no analysis, theorem, or controlled comparison quantifies this alteration relative to non-split FL. Without such grounding, the motivation for architecture-aware bias compensation remains untested and load-bearing for the novelty claim.
minor comments (2)
  1. [Abstract] The abstract asserts 'consistent outperformance' and 'extensive experiments' but supplies no quantitative metrics, error bars, or dataset-specific results; these should be summarized with key numbers in the abstract for clarity.
  2. [Method] Notation for evidential uncertainty in EA and the pairing logic in BCC should be formalized with explicit equations early in the method section to improve readability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive feedback. We address each major comment below and describe the revisions we will make to strengthen the manuscript.

read point-by-point responses

  1. Referee: [Experiments] Experiments section: The comparisons are limited to existing SFL adaptations. No ablation applies the identical EA/BCC/DTD components inside a non-split baseline such as FedAvg or FedProx (with all other hyperparameters fixed). This leaves open the possibility that reported gains arise from the reweighting and distillation mechanisms themselves rather than from any split-specific representational change, directly undermining the architecture-aware premise stated in the abstract and introduction.

    Authors: We agree that this ablation is necessary to isolate the contribution of the split architecture. In the revised manuscript we will add experiments that apply the identical EA, BCC, and DTD components to non-split baselines (FedAvg and FedProx) under the same non-IID partitions and hyperparameter settings. These results will be reported alongside the existing SFL comparisons to clarify whether the observed gains are architecture-specific. revision: yes

  2. Referee: [Introduction and §3] Introduction and §3: The observation that 'the split architecture of SFL inherently alters how client information is represented and coordinated' is presented as the foundation for the three techniques, yet no analysis, theorem, or controlled comparison quantifies this alteration relative to non-split FL. Without such grounding, the motivation for architecture-aware bias compensation remains untested and load-bearing for the novelty claim.

    Authors: The claim is grounded in the structural properties of SFL, particularly the cut-layer representation and the decoupling of client and server models, which enable techniques such as BCC that operate directly on split-layer features. While we do not provide a formal theorem, we will expand the discussion in Section 3 with a more detailed qualitative analysis of these structural differences and add a controlled comparison (split versus non-split) in the experiments to quantify their effect on bias compensation. This will better support the architecture-aware motivation. revision: yes

Circularity Check

0 steps flagged

Derivation chain self-contained; no reduction to fitted inputs or self-citations

full rationale

The paper grounds its framework in an architectural observation about SFL (split models alter client information representation and coordination) and then introduces three new components—Evidential Aggregation, Bias-Compensated Collaboration, and Dual-Teacher Distillation—explicitly derived from that observation. No equations, parameter-fitting steps, or self-citations are presented that would make any claimed prediction or result equivalent to its own inputs by construction. The work reports independent experimental comparisons against existing SFL adaptations on five datasets, satisfying the criterion for a self-contained derivation against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 3 invented entities

The abstract provides no explicit free parameters, mathematical axioms, or invented physical entities. The central claim rests on the effectiveness of three newly introduced algorithmic components whose internal parameters and assumptions are not detailed in the available text.

axioms (1)
  • domain assumption Non-IID data distributions cause biased optimization and unstable convergence in split federated learning
    Stated directly in the abstract as the core problem the framework addresses.
invented entities (3)
  • Evidential Aggregation no independent evidence
    purpose: Fine-grained reweighting of client contributions based on evidential uncertainty
    New aggregation method introduced to prevent biased local data from dominating global updates.
  • Bias-Compensated Collaboration no independent evidence
    purpose: Align split-layer representations by pairing complementary clients
    New collaboration mechanism to reduce distributional skew.
  • Dual-Teacher Distillation no independent evidence
    purpose: Synchronize knowledge between decoupled client and server models for independent local inference
    Distillation technique incorporated to enable independent local inference.

pith-pipeline@v0.9.0 · 5746 in / 1483 out tokens · 56550 ms · 2026-05-20T14:23:32.084302+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

79 extracted references · 79 canonical work pages · 2 internal anchors

  1. [1]

    Journal of the ACM (JACM) , volume=

    Linear-time approximation for maximum weight matching , author=. Journal of the ACM (JACM) , volume=. 2014 , publisher=

    work page 2014

  2. [2]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    HealSplit: Towards Self-Healing through Adversarial Distillation in Split Federated Learning , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    work page

  3. [3]

    IEEE Transactions on Mobile Computing , year=

    HASFL: Heterogeneity-aware split federated learning over edge computing systems , author=. IEEE Transactions on Mobile Computing , year=

    work page

  4. [4]

    Proceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence , pages=

    Heterogeneous federated learning with scalable server mixture-of-experts , author=. Proceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence , pages=

    work page

  5. [5]

    arXiv preprint arXiv:2511.11240 , year=

    HealSplit: Towards Self-Healing through Adversarial Distillation in Split Federated Learning , author=. arXiv preprint arXiv:2511.11240 , year=

    work page arXiv

  6. [6]

    arXiv preprint arXiv:2012.13995 , year=

    Fltrust: Byzantine-robust federated learning via trust bootstrapping , author=. arXiv preprint arXiv:2012.13995 , year=

    work page arXiv 2012

  7. [7]

    IEEE signal processing magazine , volume=

    The mnist database of handwritten digit images for machine learning research [best of the web] , author=. IEEE signal processing magazine , volume=. 2012 , publisher=

    work page 2012

  8. [8]

    Proceedings of the AAAI conference on artificial intelligence , volume=

    Reliable conflictive multi-view learning , author=. Proceedings of the AAAI conference on artificial intelligence , volume=

    work page

  9. [9]

    IEEE Transactions on Mobile Computing , year=

    From Non-IID to IID: Mobility-aware Hierarchical Federated Learning with Client-Edge Association Control , author=. IEEE Transactions on Mobile Computing , year=

    work page

  10. [10]

    arXiv preprint arXiv:2411.12377 , year=

    Non-IID data in federated learning: A survey with taxonomy, metrics, methods, frameworks and future directions , author=. arXiv preprint arXiv:2411.12377 , year=

    work page arXiv

  11. [11]

    Journal of the Franklin Institute , volume=

    The jensen-shannon divergence , author=. Journal of the Franklin Institute , volume=. 1997 , publisher=

    work page 1997

  12. [12]

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

    Deep residual learning for image recognition , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

    work page

  13. [13]

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

    Densely connected convolutional networks , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

    work page

  14. [14]

    2009 , publisher=

    Learning multiple layers of features from tiny images , author=. 2009 , publisher=

    work page 2009

  15. [15]

    Advances in Neural Information Processing Systems , volume=

    Convergence analysis of split federated learning on heterogeneous data , author=. Advances in Neural Information Processing Systems , volume=

    work page

  16. [16]

    IEEE Transactions on Network Science and Engineering , year=

    FedVaccine: Robust Federated Learning in Noisy and Non-IID Wireless Network Environments , author=. IEEE Transactions on Network Science and Engineering , year=

    work page

  17. [17]

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

    Exponential moving average normalization for self-supervised and semi-supervised learning , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    work page

  18. [18]

    Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms

    Fashion-mnist: a novel image dataset for benchmarking machine learning algorithms , author=. arXiv preprint arXiv:1708.07747 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  19. [19]

    Scientific data , volume=

    The HAM10000 dataset, a large collection of multi-source dermatoscopic images of common pigmented skin lesions , author=. Scientific data , volume=. 2018 , publisher=

    work page 2018

  20. [20]

    Proceedings of Machine learning and systems , volume=

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

    work page

  21. [21]

    Proceedings of the AAAI conference on artificial intelligence , volume=

    Federated learning with extremely noisy clients via negative distillation , author=. Proceedings of the AAAI conference on artificial intelligence , volume=

    work page

  22. [22]

    arXiv preprint arXiv:2111.04263 , year=

    Federated learning based on dynamic regularization , author=. arXiv preprint arXiv:2111.04263 , year=

    work page arXiv

  23. [23]

    Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    Fortifying Federated Learning Towards Trustworthiness via Auditable Data Valuation and Verifiable Client Contribution , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    work page

  24. [24]

    International conference on machine learning , pages=

    Scaffold: Stochastic controlled averaging for federated learning , author=. International conference on machine learning , pages=. 2020 , organization=

    work page 2020

  25. [25]

    IEEE Internet of Things Journal , volume=

    Federated learning with non-iid data: A survey , author=. IEEE Internet of Things Journal , volume=. 2024 , publisher=

    work page 2024

  26. [26]

    IEEE Transactions on Information Forensics and Security , year=

    Maximizing uncertainty for federated learning via bayesian optimisation-based model poisoning , author=. IEEE Transactions on Information Forensics and Security , year=

    work page

  27. [27]

    Advances in Neural Information Processing Systems , volume=

    Evidential Mixture Machines: Deciphering Multi-Label Correlations for Active Learning Sensitivity , author=. Advances in Neural Information Processing Systems , volume=

    work page

  28. [28]

    Advances in Neural Information Processing Systems , volume=

    Flamby: Datasets and benchmarks for cross-silo federated learning in realistic healthcare settings , author=. Advances in Neural Information Processing Systems , volume=

    work page

  29. [29]

    Advances in Neural Information Processing Systems , volume=

    Hyper-opinion evidential deep learning for out-of-distribution detection , author=. Advances in Neural Information Processing Systems , volume=

    work page

  30. [30]

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

    Deep neural networks are easily fooled: High confidence predictions for unrecognizable images , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

    work page

  31. [31]

    arXiv preprint arXiv:2302.13824 , year=

    Dirichlet-based uncertainty calibration for active domain adaptation , author=. arXiv preprint arXiv:2302.13824 , year=

    work page arXiv

  32. [32]

    Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    A simple data augmentation for feature distribution skewed federated learning , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    work page

  33. [33]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Post-hoc uncertainty learning using a dirichlet meta-model , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    work page

  34. [34]

    IEEE Transactions on Dependable and Secure Computing , year=

    Finding the PISTE: Towards understanding privacy leaks in vertical federated learning systems , author=. IEEE Transactions on Dependable and Secure Computing , year=

    work page

  35. [35]

    European Conference on Computer Vision , pages=

    Select and distill: Selective dual-teacher knowledge transfer for continual learning on vision-language models , author=. European Conference on Computer Vision , pages=. 2024 , organization=

    work page 2024

  36. [36]

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

    Think twice before selection: Federated evidential active learning for medical image analysis with domain shifts , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    work page

  37. [37]

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

    Rscfed: Random sampling consensus federated semi-supervised learning , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    work page

  38. [38]

    IEEE Transactions on Information Forensics and Security , year=

    Evaluating security and robustness for split federated learning against poisoning attacks , author=. IEEE Transactions on Information Forensics and Security , year=

    work page

  39. [39]

    Advances in neural information processing systems , volume=

    Evidential deep learning to quantify classification uncertainty , author=. Advances in neural information processing systems , volume=

    work page

  40. [40]

    Proceedings of the 30th Annual International Conference on Mobile Computing and Networking , pages=

    Parallelsfl: A novel split federated learning framework tackling heterogeneity issues , author=. Proceedings of the 30th Annual International Conference on Mobile Computing and Networking , pages=

    work page

  41. [41]

    Advances in Neural Information Processing Systems , volume=

    Confusion-resistant federated learning via diffusion-based data harmonization on non-IID data , author=. Advances in Neural Information Processing Systems , volume=

    work page

  42. [42]

    2024 IEEE 40th International Conference on Data Engineering (ICDE) , pages=

    MergeSFL: Split federated learning with feature merging and batch size regulation , author=. 2024 IEEE 40th International Conference on Data Engineering (ICDE) , pages=. 2024 , organization=

    work page 2024

  43. [43]

    Operations Research , volume=

    Greedy algorithm for multiway matching with bounded regret , author=. Operations Research , volume=. 2024 , publisher=

    work page 2024

  44. [44]

    IEEE Transactions on Mobile Computing , year=

    Minimization of the Training Makespan in Hybrid Federated Split Learning , author=. IEEE Transactions on Mobile Computing , year=

    work page

  45. [45]

    Scientific Reports , volume=

    Privacy-preserving federated learning for collaborative medical data mining in multi-institutional settings , author=. Scientific Reports , volume=. 2025 , publisher=

    work page 2025

  46. [46]

    Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    Federated Learning with Domain Shift Eraser , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    work page

  47. [47]

    IEEE Transactions on Dependable and Secure Computing , year=

    Split learning without local weight sharing to enhance client-side data privacy , author=. IEEE Transactions on Dependable and Secure Computing , year=

    work page

  48. [48]

    IEEE Transactions on Network Science and Engineering , volume=

    Federated split learning via mutual knowledge distillation , author=. IEEE Transactions on Network Science and Engineering , volume=. 2024 , publisher=

    work page 2024

  49. [49]

    2018 , publisher=

    Subjective Logic: A formalism for reasoning under uncertainty , author=. 2018 , publisher=

    work page 2018

  50. [50]

    Journal of the Royal Statistical Society: Series B (Methodological) , volume=

    A generalization of Bayesian inference , author=. Journal of the Royal Statistical Society: Series B (Methodological) , volume=. 1968 , publisher=

    work page 1968

  51. [51]

    1976 , publisher=

    A mathematical theory of evidence , author=. 1976 , publisher=

    work page 1976

  52. [52]

    2020 , publisher=

    A mathematical theory of evidence , author=. 2020 , publisher=

    work page 2020

  53. [53]

    The Twelfth International Conference on Learning Representations , year=

    Fedcda: Federated learning with cross-rounds divergence-aware aggregation , author=. The Twelfth International Conference on Learning Representations , year=

    work page

  54. [54]

    Advances in Neural Information Processing Systems , volume=

    Compact language models via pruning and knowledge distillation , author=. Advances in Neural Information Processing Systems , volume=

    work page

  55. [55]

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

    Logit standardization in knowledge distillation , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    work page

  56. [56]

    Advances in Neural Information Processing Systems , volume=

    FedGMKD: An efficient prototype federated learning framework through knowledge distillation and discrepancy-aware aggregation , author=. Advances in Neural Information Processing Systems , volume=

    work page

  57. [57]

    IEEE Transactions on Pattern Analysis and Machine Intelligence , volume=

    Learning from human educational wisdom: A student-centered knowledge distillation method , author=. IEEE Transactions on Pattern Analysis and Machine Intelligence , volume=. 2024 , publisher=

    work page 2024

  58. [58]

    2024 IEEE 40th International Conference on Data Engineering (ICDE) , pages=

    Dual-teacher de-biasing distillation framework for multi-domain fake news detection , author=. 2024 IEEE 40th International Conference on Data Engineering (ICDE) , pages=. 2024 , organization=

    work page 2024

  59. [59]

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

    Frequency attention for knowledge distillation , author=. Proceedings of the IEEE/CVF winter conference on applications of computer vision , pages=

    work page

  60. [60]

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

    Small scale data-free knowledge distillation , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    work page

  61. [61]

    Advances in neural information processing systems , volume=

    Group knowledge transfer: Federated learning of large cnns at the edge , author=. Advances in neural information processing systems , volume=

    work page

  62. [62]

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

    An aggregation-free federated learning for tackling data heterogeneity , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    work page

  63. [63]

    IEEE Transactions on Pattern Analysis and Machine Intelligence , volume=

    Federated learning for generalization, robustness, fairness: A survey and benchmark , author=. IEEE Transactions on Pattern Analysis and Machine Intelligence , volume=. 2024 , publisher=

    work page 2024

  64. [64]

    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=

    work page

  65. [65]

    Nature communications , volume=

    Introducing edge intelligence to smart meters via federated split learning , author=. Nature communications , volume=. 2024 , publisher=

    work page 2024

  66. [66]

    IEEE/ACM Transactions on Networking , volume=

    Accelerating federated learning with data and model parallelism in edge computing , author=. IEEE/ACM Transactions on Networking , volume=. 2023 , publisher=

    work page 2023

  67. [67]

    Proceedings of the Twentieth European Conference on Computer Systems , pages=

    Hourglass: Enabling Efficient Split Federated Learning with Data Parallelism , author=. Proceedings of the Twentieth European Conference on Computer Systems , pages=

    work page

  68. [68]

    Proceedings of the ACM Web Conference 2022 , pages=

    Locfedmix-sl: Localize, federate, and mix for improved scalability, convergence, and latency in split learning , author=. Proceedings of the ACM Web Conference 2022 , pages=

    work page 2022

  69. [69]

    IEEE Transactions on Mobile Computing , year=

    Hierarchical split federated learning: Convergence analysis and system optimization , author=. IEEE Transactions on Mobile Computing , year=

    work page

  70. [70]

    Proceedings of the AAAI conference on artificial intelligence , volume=

    Splitfed: When federated learning meets split learning , author=. Proceedings of the AAAI conference on artificial intelligence , volume=

    work page

  71. [71]

    RESFL: An Uncertainty-Aware Framework for Responsible Federated Learning by Balancing Privacy, Fairness and Utility

    RESFL: An Uncertainty-Aware Framework for Responsible Federated Learning by Balancing Privacy, Fairness and Utility in Autonomous Vehicles , author=. arXiv preprint arXiv:2503.16251 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  72. [72]

    Artificial intelligence and statistics , pages=

    Communication-efficient learning of deep networks from decentralized data , author=. Artificial intelligence and statistics , pages=. 2017 , organization=

    work page 2017

  73. [73]

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

    Re-thinking federated active learning based on inter-class diversity , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    work page

  74. [74]

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

    FedEvi: Improving Federated Medical Image Segmentation via Evidential Weight Aggregation , author=. International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=. 2024 , organization=

    work page 2024

  75. [75]

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

    Reliable and interpretable personalized federated learning , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    work page

  76. [76]

    IEEE transactions on pattern analysis and machine intelligence , volume=

    Trusted multi-view classification with dynamic evidential fusion , author=. IEEE transactions on pattern analysis and machine intelligence , volume=. 2022 , publisher=

    work page 2022

  77. [77]

    Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V

    Uncertainty quantification and confidence calibration in large language models: A survey , author=. Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V. 2 , pages=

    work page

  78. [78]

    IEEE Transactions on Pattern Analysis and Machine Intelligence , year=

    Revisiting essential and nonessential settings of evidential deep learning , author=. IEEE Transactions on Pattern Analysis and Machine Intelligence , year=

    work page

  79. [79]

    IEEE Transactions on Pattern Analysis and Machine Intelligence , volume=

    Evidential multi-source-free unsupervised domain adaptation , author=. IEEE Transactions on Pattern Analysis and Machine Intelligence , volume=. 2024 , publisher=

    work page 2024