Pith. sign in

REVIEW 5 major objections 4 minor 1 cited by

Handling Spatial-Temporal Data Heterogeneity for Federated Continual Learning via Tail Anchor

T0 review · 5 major / 4 minor · reviewed 2026-08-11 · deepseek-v4-flash

Pith's one-line read The paper argues that spatial-temporal catastrophic forgetting in federated continual learning can be controlled by freezing a pre-trained ViT and mixing learnable Tail Anchors with frozen output features to hold every class at a fixed…

desk verdict FedTA is a prompt-based FCL method with strong reported gains, but its client-specific input tokens undercut the paper's central claim of a shared frozen-ViT feature space. read the letter →

arxiv 2412.18355 v2 pith:FFPB4FBW submitted 2024-12-24 cs.CV cs.AIcs.LG

classification cs.CVcs.AIcs.LG
keywords federatedcontinuallearningspatial-temporaldataheterogeneitycatastrophicforgettingpre-trainedVisionTransformertailanchorinputenhancementglobalprototypeselectionknowledgeretention
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

Federated continual learning couples two sources of data shift: each client holds a different private slice of classes (spatial heterogeneity), and every client's data arrives as a sequence of new tasks (temporal heterogeneity). The paper argues that this input-level heterogeneity moves both the parameters and the output features of a model, and names those two failure modes parameter-forgetting and output-forgetting. To stop them, it proposes Federated Tail Anchor (FedTA): keep a pre-trained Vision Transformer frozen, prepend small trainable input tokens, and append trainable class-specific tail anchors to the output features so every class has a fixed target position in feature space. In experiments on CIFAR-100 and ImageNet-R, FedTA keeps accuracy high across five tasks and five clients and keeps temporal and spatial knowledge retention around 98 percent.

What carries the argument

The Tail Anchor is the load-bearing object: a set of key-value pairs, one learnable vector per class, with a key used to select the right anchor by cosine similarity to the frozen output feature. The anchor is concatenated (mixed) with the output feature to form a new feature, and a contrastive loss ties that mixed feature to a global prototype fixed by the server. Because the anchor's relative position is defined against a frozen global prototype, it can pull drifted features back to their original class location even when inputs change across clients and tasks. Input Enhancement tokens and the server-side knowledge fusion play a supporting role: they let the frozen ViT handle new classes and let the server merge heterogeneous local tokens without touching the backbone.

What would settle it

Train FedTA with clients using different pre-trained backbones (or the same backbone pre-trained on different data), so local features are not in one shared space, and check whether the server-selected global prototypes still lift final-task accuracy; a large drop would confirm that the method's gains depend on the shared-feature-space assumption rather than on the anchor mechanism itself.

Watch

Extended reading notes

Core claim

The central claim is that spatial-temporal catastrophic forgetting in federated continual learning can be reduced to a feature-position problem and solved without updating the backbone. FedTA uses a frozen pre-trained ViT shared by all clients, which removes parameter-forgetting because the feature extractor never changes. Input Enhancement adds a small set of trainable tokens to the patch embeddings so the frozen model adapts to downstream classes; Selective Input Knowledge Fusion distills these tokens on the server; Best Global Prototype Selection picks, for each class, the local prototype least similar to all others and fixes it as a global anchor. During local training, the learned Tail Anchor is mixed with the frozen output features and pulled toward that anchor by contrastive loss. The paper reports that FedTA outperforms FCL baselines, that the ablation removing the tail anchor causes the largest accuracy drop, and that t-SNE visualizations show feature positions stay stable across tasks and clients.

Load-bearing premise

All clients share one frozen pre-trained ViT whose feature space is the same for every client and every task, so anchors and prototypes computed from local features remain valid for features produced later by other clients; if that shared space drifts or was never aligned, the anchors would be attached to the wrong coordinates.

Editorial extensions

If this is right

  • Clients only communicate input-enhancement tokens and class prototypes; if FedTA holds, the backbone never needs to be sent, which cuts communication and lets heterogeneous clients share one frozen encoder.
  • Forgetting is controlled by fixing feature positions rather than replaying old data, so FedTA avoids the privacy and storage costs of replay-based FCL baselines.
  • Because the tail anchor is the component that holds features in place, any new task can be added by learning a new anchor vector, making class-incremental growth a matter of adding key-value pairs.
  • The reported gap between FedTA and the strongest baseline (e.g., final-task accuracy 89.4 versus 82.2 on CIFAR-100) is driven mostly by the tail anchor, per the ablation, so future FCL methods should treat feature-position anchoring as a first-class mechanism.

Reading between the lines

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

  • If the shared-feature-space premise is right, FedTA's recipe transfers to any frozen encoder, so the same anchor mechanism could be used for federated continual learning of text or audio by changing only the encoder.
  • A testable extension is to replace the server's lowest-similarity prototype rule with a maximum-margin rule (pick the prototype that maximizes separation from all other classes); the paper's adjacency-matrix machinery already computes the needed similarities.
  • The method's reliance on a single global pre-trained model suggests a natural stress test: apply FedTA in a setting where clients receive the same architecture but with different pre-trained weights, and measure how quickly anchor alignment degrades.
  • If output-forgetting is truly the main failure mode, then even without input enhancement, the tail anchor should retain most of the benefit; an ablation that isolates the anchor from Input Enhancement would separate the value of position-fixing from the value of adapter tokens.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 4 minor

Summary. The paper proposes FedTA, a federated continual learning (FCL) method that combines a frozen pretrained Vision Transformer (ViT) with trainable "input enhancement" tokens and trainable "tail anchors." The core idea is that, because the ViT is frozen, all clients share the same feature extractor; the method then mixes tail anchors with the frozen output features to keep class features near fixed global prototypes, thereby preventing what the authors call parameter-forgetting and output-forgetting. The server also performs selective input knowledge fusion (SIKF) on the input enhancement knowledge bases and best global prototype selection (BGPS) on uploaded local prototypes. Experiments on CIFAR-100 and ImageNet-R compare FedTA against several FL, FL+CL, and FCL baselines, reporting accuracy and two knowledge-retention metrics, plus ablations and t-SNE visualizations.

Significance. If the claims are validated, FedTA would be a useful contribution to FCL: it avoids replay data, keeps communication costs small, and reports strong gains over existing methods while using a frozen pretrained backbone. The paper also addresses an important and under-studied problem, spatial-temporal catastrophic forgetting, and proposes a mechanism that is conceptually interesting: fixing feature positions via learnable anchors that are mixed with frozen features. The use of a frozen ViT to sidestep parameter forgetting is a promising direction. The authors are also to be credited for including ablations, retention metrics, and an efficiency/privacy discussion. However, the current manuscript has substantial reproducibility and conceptual gaps: the central mixing operation is never defined mathematically, the input enhancement component appears to violate the paper's own premise of a shared feature space, and the experimental section reports no error bars or hyperparameter values. These issues must be resolved before the empirical claims can be fully assessed.

major comments (5)
  1. [Sec. 4.2, Eq. (5), Eq. (7)] The central operation of FedTA, mixing a tail anchor with the output feature to form F_TA, is never defined. The text says "it will be mixed with Fout to form a new feature F_TA" but provides no equation, no description of whether the mixing is a weighted sum, concatenation, or attention-like combination, and no statement of the resulting dimension. This makes Eqs. (5), (6), and (7) ambiguous and the method unreproducible. Please provide the exact operation, and also specify how F_TA is fed to the classifier for the cross-entropy loss in Eq. (6).
  2. [Sec. 3.3 vs. Sec. 4.1, Eq. (2)] The premise stated in Sec. 3.3 is that all clients share the same frozen pretrained ViT, "ensuring that they have the same knowledge/feature space." However, Sec. 4.1 introduces client-specific input enhancement tokens that are concatenated to the patch embeddings in Eq. (2) and then processed by the frozen ViT. Because self-attention mixes these tokens with image patches, the output feature Fout for the same image will differ across clients and across tasks, even with frozen ViT weights. Consequently, local prototypes computed via Eq. (7) live in client-dependent feature spaces, and using a global prototype selected from one client's space in the contrastive loss (Eq. (5)) on another client assumes that the IE-induced shift is negligible. The paper never measures or bounds this shift. This is load-bearing for the claim that FedTA preserves feature positions and enables a common feature space. Please address this directly, e.g., by analyzing the shift empirically or by modifying the method so that prototypes are computed in a truly shared space.
  3. [Table 1 and Sec. 5.1] The main experimental result consists of single-run accuracy numbers with no error bars, no standard deviations, and no number of seeds. Given that the reported gains over the strongest baseline (FedMGP) are sometimes modest, and that the ablation comparisons (e.g., Ours vs. Ours-w/o SIKF on ImageNet-R) are close, the reader cannot assess whether the differences are statistically meaningful. In addition, hyperparameters are not reported: the loss weights lambda1, lambda2, lambda3 in Eqs. (3) and (6), the temperature tau in Eq. (5), the threshold Th in Sec. 4.4, the knowledge base size M and selected count N in Eq. (2), and the optimizer settings are all absent. Please provide these details and multiple-seed results with error bars.
  4. [Table 1 and Sec. 5.3] The ablation for Selective Input Knowledge Fusion does not support the claim that SIKF is beneficial. On ImageNet-R, Ours-w/o SIKF achieves 80.0/80.5/81.1/82.9/81.7 across tasks, while full FedTA achieves 81.5/78.8/79.2/80.6/85.0. Thus at tasks 2, 3, and 4 the variant without SIKF is better. The text admits this "sometimes falls below" but the claim in the abstract and contributions that SIKF is a novel and effective component is not substantiated. Please either provide a more thorough analysis (e.g., with variance, additional datasets, or a corrected mechanism) or temper the claim.
  5. [Sec. 4.4, Eqs. (9) and (10)] The Best Global Prototype Selection criterion is not well justified. Eq. (9) defines M_ij via a function "dis", which is elsewhere used as distance/cosine similarity, but then states that M_ij = 1 for prototypes of the same class; this is inconsistent (if it is a distance, same-class entries should be 0; if it is a similarity, the text should say so and define the range). Moreover, Eq. (10) selects the prototype with the lowest average similarity as the global anchor, but the paper provides no argument why the least representative prototype in this sense is the best anchor point for a class. This is a core component, so please clarify the definition and provide an empirical or theoretical justification.
minor comments (4)
  1. [Fig. 1] The label "PositionDift" appears to be a typo for "Position Drift."
  2. [Eq. (3)] The notation for keys is confusing: K_ie^s, K_ie^in, and the summation over "Kie_s" are not clearly defined. Please restate the surrogate loss and specify what is being summed over.
  3. [Sec. 5.4] The privacy discussion states that "If we randomly mix Tail Anchor with features, this issue will be resolved," but random mixing would appear to destroy the method's ability to place features at fixed class-specific positions. Please clarify this statement.
  4. [Fig. 5] The t-SNE figure labels the number of input enhancement tokens and tail anchors but does not state which dataset, which task, or which classes are shown, nor whether the "same samples" are from old tasks. Please make the figure self-contained.

Circularity Check

1 steps flagged · score 4.0 of 10

Feature-position preservation is partly by construction: the contrastive loss trains features toward global prototypes that are computed from those same features, so the reported stability is the training objective itself; the core accuracy results remain externally benchmarked.

  1. self definitional [Sec. 4.2 (Tail Anchor and Eq. 5-7), Sec. 4.4 (Eq. 10), and Sec. 5.3 (visualization claim)]
    "the tail anchor of each class acts as additional parameters to manage the distance to a fixed position (global prototype of each class) in the feature space. Its main advantage is its relative position, which remains consistent regardless of changes in space and time. ... The local prototype is obtained by averaging features with tail anchors belonging to the same class, computed through P_y_i = 1/|D_y_a| Σ F_x_TA ... the server selects the prototype with the lowest average similarity ... as the global prototype ... Fig."

    The 'fixed position' used as the contrastive target in Eq. (5) is not an external ground truth: the global prototype is selected by Eq. (10) from local prototypes that are themselves averages of the already tail-anchored features F_TA (Eq. 7). The optimization loss in Eq. (6) directly trains F_TA toward those global prototypes via L_cons. Therefore the reported observation that features 'preserve' their positions and relative distances near the anchors is the training objective itself, not an independent prediction or discovery. The benchmark accuracy in Table 1 is externally evaluated and not circular, but the feature-position preservation claim is enforced by construction.

full rationale

The paper's central accuracy and knowledge-retention claims rest on benchmark comparisons against FedAvg, FedProx, GLFC, TARGET, MFCL, and other external baselines, so the main experimental result is not circular. However, one stated contribution - that FedTA 'effectively preserves the relative positions of features' - reduces by construction to the method's own objective: tail anchors are trained with a contrastive loss that pulls features toward global prototypes, and those global prototypes are computed by averaging the very tail-anchored features they anchor. This makes the feature-space stability observation self-referential. The paper also cites prior work by the same authors for the ST-CF concept and metrics ([35], [38]), but the metrics are explicitly defined and the accuracy comparisons are independent, so these self-citations are not load-bearing circularity. The additional concern that client-specific Input Enhancement tokens violate the paper's premise of a shared frozen-ViT feature space is an internal consistency / correctness risk rather than a circularity pattern, and does not by itself raise the circularity score further.

Assumptions & free parameters 6 free parameters · 5 assumptions · 1 invented entities

The central claim rests on several hand-chosen hyperparameters, a surrogate-dataset assumption, and an underspecified mixing operation. The Tail Anchor is the paper's primary invented mechanism and currently has no evidence outside these experiments.

free parameters (6)
  • Number of tail anchors and dimension = 100 x 768
    Manually set in Sec 5.4; not fitted to data but requires selection for each class count.
  • Input enhancement base size M and selected count N = 10 and 10
    Set to 10 in Sec 5.4; controls capacity of the knowledge base.
  • Loss weights lambda1, lambda2, lambda3 = not reported
    Weights in Eqs (3) and (6) balance classification, contrastive, and key-similarity terms; values are not given in the paper.
  • Temperature tau in contrastive loss = not reported
    Controls tolerance in Eq (5); not specified.
  • Threshold Th for global prototype fixation = not reported
    In Sec 4.4, the global prototype is fixed when average similarity falls below Th; value not given.
  • Optimization hyperparameters (learning rate, epochs, batch size) = not reported
    Training details are absent, preventing replication.
assumptions (5)
  • domain assumption Frozen pre-trained ViT provides a shared and stable feature space across clients and tasks.
    Sec 3.3: 'clients share the same pre-trained model, ensuring that they have the same knowledge/feature space'; the whole method depends on this.
  • domain assumption The server has access to a small surrogate dataset Ds for distillation.
    Sec 4.3: 'We follow a common setting, which allows the server to possess a small-scale surrogate dataset'; may be unavailable in strict FL.
  • domain assumption Adding learnable tokens to input embeddings is sufficient to adapt a frozen ViT to downstream tasks for each client.
    Sec 4.1: Input Enhancement concatenates selected IE tokens to embedding E; this presumes the frozen backbone needs no updates.
  • ad hoc to paper The global prototype with the lowest average similarity is the best anchor point for each class.
    Sec 4.4: heuristic selection rule; no justification beyond empirical results.
  • ad hoc to paper Mixing tail anchors with output features through an unspecified operation controls feature position without disrupting the classification of the frozen model.
    Sec 4.2: the mixing is never defined with an equation; the method relies on this underspecified operation.
invented entities (1)
  • Tail Anchor (TA)
    purpose: learnable key-value vectors that, when mixed with frozen output features, fix class positions in feature space
    Introduced in this paper as a new mechanism; no external validation beyond its empirical use here.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Handling Spatial-Temporal Data Heterogeneity for Federated Continual Learning via Tail Anchor." pith.science (2026). https://pith.science/paper/FFPB4FBW

@misc{pith2026241218355,
  author       = {Pith},
  title        = {Pith review of: Handling Spatial-Temporal Data Heterogeneity for Federated Continual Learning via Tail Anchor},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FFPB4FBW}},
  note         = {Machine review of arXiv:2412.18355}
}
read the original abstract

Federated continual learning (FCL) allows each client to continually update its knowledge from task streams, enhancing the applicability of federated learning in real-world scenarios. However, FCL needs to address not only spatial data heterogeneity between clients but also temporal data heterogeneity between tasks. In this paper, empirical experiments demonstrate that such input-level heterogeneity significantly affects the model's internal parameters and outputs, leading to severe spatial-temporal catastrophic forgetting of local and previous knowledge. To this end, we propose Federated Tail Anchor (FedTA) to mix trainable Tail Anchor with the frozen output features to adjust their position in the feature space, thereby overcoming parameter-forgetting and output-forgetting. Three novel components are also included: Input Enhancement for improving the performance of pre-trained models on downstream tasks; Selective Input Knowledge Fusion for fusion of heterogeneous local knowledge on the server; and Best Global Prototype Selection for finding the best anchor point for each class in the feature space. Extensive experiments demonstrate that FedTA not only outperforms existing FCL methods but also effectively preserves the relative positions of features.

Figures

Figures reproduced from arXiv: 2412.18355 by the authors.

Figure 1
Figure 1. Illustration of FCL, the negative impact of spatial-temporal data heterogeneity and the intuition of Tail Anchor. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Illustrations of the negative impact of spatial-temporal data heterogeneity on the feature extractor and feature space. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. An overview of FedTA. Local training is a two-stage training process. The first stage involves adding [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Knowledge retention on different dataset. [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: T-SNE for position changes of features corresponding to the same samples after FCL. [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Unleashing the Power of Continual Learning on Non-Centralized Devices: A Survey

    cs.LG 2024-12 conditional novelty 4.0 of 10

    A review of non-centralized continual learning that taxonomizes data-, model-, and device-level methods and benchmarks twelve federated continual learning methods on six datasets.

Reference graph

Works this paper leans on

44 extracted references · 37 canonical work pages · cited by 1 Pith paper

  1. [1]

    A data-free ap- proach to mitigate catastrophic forgetting in federated class incremental learning for vision tasks

    Sara Babakniya, Zalan Fabian, Chaoyang He, Mahdi Soltanolkotabi, and Salman Avestimehr. A data-free ap- proach to mitigate catastrophic forgetting in federated class incremental learning for vision tasks. Advances in Neural Information Processing Systems, 36, 2024. 1, 3, 7

  2. [2]

    Feder- ated learning with client subsampling, data heterogeneity, and unbounded smoothness: A new algorithm and lower bounds

    Michael Crawshaw, Yajie Bao, and Mingrui Liu. Feder- ated learning with client subsampling, data heterogeneity, and unbounded smoothness: A new algorithm and lower bounds. Advances in Neural Information Processing Sys- tems, 36, 2024. 2

  3. [3]

    Federated class-incremental learn- ing

    Jiahua Dong, Lixu Wang, Zhen Fang, Gan Sun, Shichao Xu, Xiao Wang, and Qi Zhu. Federated class-incremental learn- ing. In Proceedings of the IEEE/CVF conference on com- puter vision and pattern recognition , pages 10164–10173,

  4. [4]

    An image is worth 16x16 words: Trans- formers for image recognition at scale

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Syl- vain Gelly, et al. An image is worth 16x16 words: Trans- formers for image recognition at scale. arXiv preprint arXiv:2010.11929, 2020. 2, 5, 7

  5. [5]

    Continual learning with transformers for image classification

    Beyza Ermis, Giovanni Zappella, Martin Wistuba, Aditya Rawal, and C ´edric Archambeau. Continual learning with transformers for image classification. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 3774–3781, 2022. 2

  6. [6]

    Ten Challenging Problems in Federated Foundation Models

    Tao Fan, Hanlin Gu, Xuemei Cao, Chee Seng Chan, Qian Chen, Yiqiang Chen, Yihui Feng, Yang Gu, Jiaxiang Geng, Bing Luo, et al. Ten challenging problems in federated foun- dation models. arXiv preprint arXiv:2502.12176, 2025. 1, 2

  7. [7]

    Fed- prok: Trustworthy federated class-incremental learning via prototypical feature knowledge transfer

    Xin Gao, Xin Yang, Hao Yu, Yan Kang, and Tianrui Li. Fed- prok: Trustworthy federated class-incremental learning via prototypical feature knowledge transfer. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 4205–4214, 2024. 1

  8. [8]

    Embracing change: Continual learning in deep neural networks

    Raia Hadsell, Dushyant Rao, Andrei A Rusu, and Razvan Pascanu. Embracing change: Continual learning in deep neural networks. Trends in cognitive sciences, 24(12):1028– 1040, 2020. 1

Show all 44 references
  1. [9]

    The many faces of robust- ness: A critical analysis of out-of-distribution generalization

    Dan Hendrycks, Steven Basart, Norman Mu, Saurav Kada- vath, Frank Wang, Evan Dorundo, Rahul Desai, Tyler Zhu, Samyak Parajuli, Mike Guo, et al. The many faces of robust- ness: A critical analysis of out-of-distribution generalization. In Proceedings of the IEEE/CVF internation...

  2. [10]

    Learn from others and be yourself in heterogeneous federated learning

    Wenke Huang, Mang Ye, and Bo Du. Learn from others and be yourself in heterogeneous federated learning. In Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 10143–10153, 2022. 6

  3. [11]

    Grounding foundation models through federated transfer learning: A general framework

    Yan Kang, Tao Fan, Hanlin Gu, Lixin Fan, and Qiang Yang. Grounding foundation models through federated transfer learning: A general framework. arXiv preprint arXiv:2311.17431, 2023. 2

  4. [12]

    Navigating data heterogeneity in fed- erated learning: a semi-supervised federated object detec- tion

    Taehyeon Kim, Eric Lin, Junu Lee, Christian Lau, and Vaikkunth Mugunthan. Navigating data heterogeneity in fed- erated learning: a semi-supervised federated object detec- tion. In Thirty-seventh Conference on Neural Information Processing Systems, 2023. 2

  5. [13]

    Overcoming catastrophic forgetting in neu- ral networks

    James Kirkpatrick, Razvan Pascanu, Neil Rabinowitz, Joel Veness, Guillaume Desjardins, Andrei A Rusu, Kieran Milan, John Quan, Tiago Ramalho, Agnieszka Grabska- Barwinska, et al. Overcoming catastrophic forgetting in neu- ral networks. Proceedings of the national academy of sc...

  6. [14]

    Learning multiple layers of features from tiny images

    Alex Krizhevsky, Geoffrey Hinton, et al. Learning multiple layers of features from tiny images. 2009. 6

  7. [15]

    Fac- ing spatiotemporal heterogeneity: A unified federated con- tinual learning framework with self-challenge rehearsal for industrial monitoring tasks

    Baoxue Li, Pengyu Song, Chunhui Zhao, and Min Xie. Fac- ing spatiotemporal heterogeneity: A unified federated con- tinual learning framework with self-challenge rehearsal for industrial monitoring tasks. Knowledge-Based Systems, 289: 111491, 2024. 3

  8. [16]

    Fed- erated learning on non-iid data silos: An experimental study

    Qinbin Li, Yiqun Diao, Quan Chen, and Bingsheng He. Fed- erated learning on non-iid data silos: An experimental study. In 2022 IEEE 38th international conference on data engi- neering (ICDE), pages 965–978. IEEE, 2022. 2

  9. [17]

    Federated optimiza- tion in heterogeneous networks

    Tian Li, Anit Kumar Sahu, Manzil Zaheer, Maziar Sanjabi, Ameet Talwalkar, and Virginia Smith. Federated optimiza- tion in heterogeneous networks. Proceedings of Machine learning and systems, 2:429–450, 2020. 7

  10. [18]

    Towards efficient re- play in federated incremental learning

    Yichen Li, Qunwei Li, Haozhao Wang, Ruixuan Li, Wen- liang Zhong, and Guannan Zhang. Towards efficient re- play in federated incremental learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 12820–12829, 2024. 3

  11. [19]

    Unleashing the power of continual learning on non-centralized devices: A survey,

    Yichen Li, Haozhao Wang, Wenchao Xu, Tianzhe Xiao, Hong Liu, Minzhu Tu, Yuying Wang, Xin Yang, Rui Zhang, Shui Yu, Song Guo, and Ruixuan Li. Unleashing the power of continual learning on non-centralized devices: A survey,

  12. [20]

    Sr-fdil: Synergistic replay for fed- erated domain-incremental learning

    Yichen Li, Wenchao Xu, Haozhao Wang, Yining Qi, Ruix- uan Li, and Song Guo. Sr-fdil: Synergistic replay for fed- erated domain-incremental learning. IEEE Transactions on Parallel and Distributed Systems, 2024. 3

  13. [21]

    Learning to prompt knowledge transfer for open-world continual learning

    Yujie Li, Xin Yang, Hao Wang, Xiangkun Wang, and Tianrui Li. Learning to prompt knowledge transfer for open-world continual learning. In Proceedings of the AAAI Conference on Artificial Intelligence, pages 13700–13708, 2024. 4

  14. [22]

    Personalized federated domain- incremental learning based on adaptive knowledge match- ing

    Yichen Li, Wenchao Xu, Haozhao Wang, Yining Qi, Jing- cai Guo, and Ruixuan Li. Personalized federated domain- incremental learning based on adaptive knowledge match- ing. In European Conference on Computer Vision , pages 127–144. Springer, 2025. 3

  15. [23]

    Learning without forgetting

    Zhizhong Li and Derek Hoiem. Learning without forgetting. IEEE transactions on pattern analysis and machine intelli- gence, 40(12):2935–2947, 2017. 1, 7

  16. [24]

    Diffusion-driven data replay: A novel approach to combat forgetting in federated class continual learning

    Jinglin Liang, Jin Zhong, Hanlin Gu, Zhongqi Lu, Xingxing Tang, Gang Dai, Shuangping Huang, Lixin Fan, and Qiang Yang. Diffusion-driven data replay: A novel approach to combat forgetting in federated class continual learning. In European Conference on Computer Vision, pages 30...

  17. [25]

    Continual federated learning based on knowledge dis- tillation

    Yuhang Ma, Zhongle Xie, Jue Wang, Ke Chen, and Lidan Shou. Continual federated learning based on knowledge dis- tillation. In IJCAI, pages 2182–2188, 2022. 6

  18. [26]

    Communication- efficient learning of deep networks from decentralized data

    Brendan McMahan, Eider Moore, Daniel Ramage, Seth Hampson, and Blaise Aguera y Arcas. Communication- efficient learning of deep networks from decentralized data. In Artificial intelligence and statistics , pages 1273–1282. PMLR, 2017. 7

  19. [27]

    Towards exemplar-free continual learning in vision transformers: an account of at- tention, functional and weight regularization

    Francesco Pelosin, Saurav Jha, Andrea Torsello, Bogdan Ra- ducanu, and Joost van de Weijer. Towards exemplar-free continual learning in vision transformers: an account of at- tention, functional and weight regularization. InProceedings of the IEEE/CVF Conference on Computer Vi...

  20. [28]

    Handling data heterogeneity via architectural de- sign for federated visual recognition

    Sara Pieri, Jose Restom, Samuel Horvath, and Hisham Cholakkal. Handling data heterogeneity via architectural de- sign for federated visual recognition. Advances in Neural Information Processing Systems, 36:4115–4136, 2023. 2

  21. [29]

    Learning representations by back-propagating er- rors

    David E Rumelhart, Geoffrey E Hinton, and Ronald J Williams. Learning representations by back-propagating er- rors. nature, 323(6088):533–536, 1986. 3

  22. [30]

    A closer look at rehearsal-free continual learning

    James Seale Smith, Junjiao Tian, Shaunak Halbe, Yen- Chang Hsu, and Zsolt Kira. A closer look at rehearsal-free continual learning. In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 2409–2419, 2023. 2

  23. [31]

    Tackling the objective inconsistency prob- lem in heterogeneous federated optimization

    Jianyu Wang, Qinghua Liu, Hao Liang, Gauri Joshi, and H Vincent Poor. Tackling the objective inconsistency prob- lem in heterogeneous federated optimization. Advances in neural information processing systems , 33:7611–7623,

  24. [32]

    Continual learning with lifelong vision trans- former

    Zhen Wang, Liu Liu, Yiqun Duan, Yajing Kong, and Dacheng Tao. Continual learning with lifelong vision trans- former. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 171–181,

  25. [33]

    Dualprompt: Complementary prompting for rehearsal-free continual learning

    Zifeng Wang, Zizhao Zhang, Sayna Ebrahimi, Ruoxi Sun, Han Zhang, Chen-Yu Lee, Xiaoqi Ren, Guolong Su, Vin- cent Perot, Jennifer Dy, et al. Dualprompt: Complementary prompting for rehearsal-free continual learning. InEuropean Conference on Computer Vision , pages 631–648. Springer,

  26. [34]

    Age-aware data selec- tion and aggregator placement for timely federated continual learning in mobile edge computing

    Zichuan Xu, Lin Wang, Weifa Liang, Qiufen Xia, Wenzheng Xu, Pan Zhou, and Omer F Rana. Age-aware data selec- tion and aggregator placement for timely federated continual learning in mobile edge computing. IEEE Transactions on Computers, 2023. 3

  27. [35]

    Federated continual learning via knowledge fu- sion: A survey

    Xin Yang, Hao Yu, Xin Gao, Hao Wang, Junbo Zhang, and Tianrui Li. Federated continual learning via knowledge fu- sion: A survey. IEEE Transactions on Knowledge and Data Engineering, 2024. 1, 2, 3, 6

  28. [36]

    Fedfed: Feature distilla- tion against data heterogeneity in federated learning

    Zhiqin Yang, Yonggang Zhang, Yu Zheng, Xinmei Tian, Hao Peng, Tongliang Liu, and Bo Han. Fedfed: Feature distilla- tion against data heterogeneity in federated learning. Ad- vances in Neural Information Processing Systems, 36, 2024. 1

  29. [37]

    Federated continual learning with weighted inter-client transfer

    Jaehong Yoon, Wonyong Jeong, Giwoong Lee, Eunho Yang, and Sung Ju Hwang. Federated continual learning with weighted inter-client transfer. In International Conference on Machine Learning, pages 12073–12086. PMLR, 2021. 1

  30. [38]

    Overcoming spatial-temporal catas- trophic forgetting for federated class-incremental learning

    Hao Yu, Xin Yang, Xin Gao, Yihui Feng, Hao Wang, Yan Kang, and Tianrui Li. Overcoming spatial-temporal catas- trophic forgetting for federated class-incremental learning. In ACM Multimedia 2024, 2024. 3

  31. [39]

    Personalized federated continual learning via multi-granularity prompt

    Hao Yu, Xin Yang, Xin Gao, Yan Kang, Hao Wang, Junbo Zhang, and Tianrui Li. Personalized federated continual learning via multi-granularity prompt. In Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, pages 4023–4034, 2024. 7

  32. [40]

    Target: Federated class-continual learning via exemplar-free distillation

    Jie Zhang, Chen Chen, Weiming Zhuang, and Lingjuan Lyu. Target: Federated class-continual learning via exemplar-free distillation. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 4782–4793, 2023. 3, 7

  33. [41]

    Cross-fcl: Toward a cross-edge federated contin- ual learning framework in mobile edge computing systems

    Zhouyangzi Zhang, Bin Guo, Wen Sun, Yan Liu, and Zhi- wen Yu. Cross-fcl: Toward a cross-edge federated contin- ual learning framework in mobile edge computing systems. IEEE Transactions on Mobile Computing , 23(1):313–326,

  34. [42]

    Federated learning with non-iid data

    Yue Zhao, Meng Li, Liangzhen Lai, Naveen Suda, Damon Civin, and Vikas Chandra. Federated learning with non-iid data. arXiv preprint arXiv:1806.00582, 2018. 2

  35. [43]

    Flee: A hierarchical federated learning framework for distributed deep neural network over cloud, edge, and end device

    Zhengyi Zhong, Weidong Bao, Ji Wang, Xiaomin Zhu, and Xiongtao Zhang. Flee: A hierarchical federated learning framework for distributed deep neural network over cloud, edge, and end device. ACM Transactions on Intelligent Sys- tems and Technology (TIST), 13(5):1–24, 2022. 3

  36. [44]

    Semi-hfl: semi-supervised federated learning for heterogeneous devices

    Zhengyi Zhong, Ji Wang, Weidong Bao, Jingxuan Zhou, Xi- aomin Zhu, and Xiongtao Zhang. Semi-hfl: semi-supervised federated learning for heterogeneous devices. Complex & Intelligent Systems, 9(2):1995–2017, 2023. 1

Pith tools

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