Pith. sign in

REVIEW 4 major objections 5 minor 47 references

Towards Macro-AUC oriented Imbalanced Multi-Label Continual Learning

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

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The paper claims a reweighted, label-distribution-aware margin loss plus a count-preserving memory update improves Macro-AUC in imbalanced multi-label continual learning, with the first generalization bounds for this setting.

desk verdict The empirical recipe (RLDAM + WRU) is promising, but the paper's theoretical contribution rests on a false Lipschitz assumption and should not be taken as established. read the letter →

arxiv 2412.18231 v1 pith:UYICLL3S submitted 2024-12-24 cs.LG

Yan Zhang , Guoqiang Wu , Bingzheng Wang , Teng Pang , Haoliang Sun , Yilong Yin This is my paper · ORCID
classification cs.LG MSC 68T0568Q32
keywords multi-labelcontinuallearningMacro-AUCclassimbalancereplay-basedmemorylabel-distribution-awaremarginreweightedunivariatelossgeneralizationboundclass-incremental
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper tries to establish that Macro-AUC, a suitable measure when multi-label classes are imbalanced, can be optimized in continual learning by a replay-based method built from two ingredients: a loss that combines per-class margin control with positive/negative reweighting, and a memory-update rule that preserves each class's positive and negative counts in the buffer. It argues that this combination gives better generalization guarantees for Macro-AUC than the existing reweighted univariate loss, and it claims these bounds are the first theoretical generalization analyses in multi-label continual learning. If the arguments hold, the method would be a practical recipe for keeping high overall multi-label accuracy under class-incremental training, and a template for theory-guided loss and memory design.

What carries the argument

The central object is the RLDAM loss, a univariate per-label surrogate that avoids pairwise computation while inheriting reweighting by reciprocal class counts and label-distribution-aware margins $\Delta_k = \lambda/|D_k^+|^{1/4}$. The theoretical core is a fractional Rademacher complexity bound for a kernel hypothesis class, a contraction inequality that reduces loss-space complexity to hypothesis-space complexity, and a margin-optimality calculation that sets margins inversely with the fourth root of positive-class size. WRU is the second mechanism: a greedy selection that keeps $|D_k^+|/|D_k^-|$ in memory close to the task's ratio, plus explicit storage of those counts, so that the replay risk inherits the original reweighting factors.

What would settle it

Compute the Lipschitz constant of $\ell(z-\Delta)=\max(0,1-(z-\Delta))$ directly: for any shift $\Delta$, the function has slope 0 or $-1$, so its Lipschitz constant is 1, not $1/\Delta$. Replacing $\rho=1/\Delta$ in Theorem 2 with $\rho=1$ removes the advertised benefit of choosing larger margins; checking whether the reported gains persist when the theory is recomputed with the true constant would settle the claim.

Watch

Extended reading notes

Core claim

The core claim is that the RLDAM loss, defined as $\ell(f_k(x^+) - \Delta_k^+) + \ell(-f_k(x^-) - \Delta_k^-)$ with per-class margins, yields a Macro-AUC generalization bound whose complexity term scales with $(\rho_k^+ + \rho_k^-)$ and can be made smaller than the corresponding term of the RU loss by choosing class-aware margins. In the continual setting, the paper claims that combining RLDAM with WRU, which stores samples so that each class's stored positive/negative ratio matches the original task, makes the replay risk an approximate unbiased estimator of the task risk and keeps the imbalance factors in the bound equal to joint-training values. The main theorem decomposes the total risk into task complexity, empirical risk, forgetting gap, and cross-task discrepancy, and the paper argues this decomposition supports both the loss and the memory-updating strategy.

Load-bearing premise

The whole margin-based improvement rests on the assumption that shifting the hinge loss by a margin $\Delta$ makes it $(1/\Delta)$-Lipschitz; in fact a shifted hinge loss is still 1-Lipschitz for any shift, so the margin-dependent shrinking of the bound is not established.

Editorial extensions

If this is right

  • On the three benchmarks considered, the full method reaches Macro-AUC of 88.69, 77.93, and 79.77, above the compared replay baselines, so practitioners could adopt RLDAM plus WRU as a drop-in replay loss and memory rule.
  • The loss alone improves Macro-AUC in batch multi-label learning over BCE and RU on the same datasets, so it can be used outside continual learning.
  • The bound suggests margins should grow as $|D_k^+|$ shrinks, making the loss a principled choice for any imbalanced multi-label dataset, not only continual streams.
  • Under WRU, memory maintains the original per-class positive/negative ratios, so replay training should not drift from joint-training imbalance factors.

Reading between the lines

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

  • If the shifted-hinge Lipschitz assumption is corrected to the true constant 1, the theoretical advantage of larger margins may disappear; empirical gains would then need to justify themselves without the bound.
  • The margin schedule $\Delta \propto |D^+|^{-1/4}$ could be turned into an online rule: recompute margins from streaming positive counts, making the method applicable when task boundaries are unknown.
  • RLDAM is a univariate surrogate, so it could be combined with regularization- or architecture-based continual learning frameworks without pairwise sampling; the paper's appendix experiment with EWC hints in that direction.
  • Applying the same reweighting-plus-margin idea to other multi-label metrics, such as F1 or mAP, would test whether the intuition generalizes beyond Macro-AUC.
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

4 major / 5 minor

Summary. The paper addresses Macro-AUC optimization in multi-label continual learning (MLCL). It proposes RLDAM, a reweighted label-distribution-aware margin loss that combines the reweighted univariate (RU) loss of Wu, Li, and Yin (2023) with the LDAM margin idea of Cao et al. (2019), and WRU, a memory-updating strategy intended to preserve class-wise positive/negative ratios in the replay buffer. The authors claim the first theoretical generalization analyses for MLCL, with batch and continual bounds showing superiority over RU, and they report experiments on C-PASCAL-VOC, C-MSCOCO, and C-NUS-WIDE.

Significance. If the theoretical claims were correct, the paper would make a useful contribution: it targets an important metric (Macro-AUC) in an underexplored setting, proposes a concrete and simple method, releases code, and includes ablation and memory-size studies. The empirical results are suggestive, with large gains over several baselines in Table 1. However, the central theoretical premise is invalid as stated: Assumption 2(3) misstates the Lipschitz constant of a shifted hinge loss, and this error propagates into Theorem 2, Proposition 1, and the MLCL bound. The claimed margin-dependent generalization guarantee and the derived optimal margins therefore do not follow. The MLCL bound also does not specifically depend on WRU. These are load-bearing issues for the paper's main contribution as framed.

major comments (4)
  1. [Section 4, Assumption 2(3)] The statement that the shifted hinge losses ℓ(z − Δ_k^+) and ℓ(z − Δ_k^-) are (1/Δ_k^+)- and (1/Δ_k^-)-Lipschitz is false. For ℓ(z) = max(0, 1 − z), the function z ↦ ℓ(z − Δ) has derivative 0 or −1 wherever differentiable, so its Lipschitz constant is exactly 1 for every Δ > 0. The footnote to Assumption 2(3) even acknowledges that the hinge loss is 1-Lipschitz. This error is load-bearing: the margin-dependent ρ terms enter Theorem 2 through Lemma 2's contraction inequality, and with ρ_k^+ = ρ_k^- = 1 the complexity term no longer depends on the margins. Consequently, the claimed superiority of the RLDAM bound over the RU bound of Wu et al. (2023) is not established.
  2. [Section 4, Proposition 1] Even within the paper's own algebra, Proposition 1 does not derive the claimed optimal margins. The minimization is over 2Λr(1/Δ_1 √(1/|D_1^+|) + 1/Δ_2 √(1/|D_2^+|)) subject to Δ_1 + Δ_2 = β, and the reported minimizer is Δ_1^* = β|D_2^+|^{1/4}/(|D_1^+|^{1/4}+|D_2^+|^{1/4}). The proposition then asserts that the optimal margins are Δ_k^* = λ/|D_k^+|^{1/4}, which is not equivalent to the derived expression for any constant λ. The further extension to more than two labels and to the equal-margin conclusion Δ_+ = Δ_- for single-label tasks is asserted without proof. Since this margin schedule is used in the experiments, the schedule lacks theoretical justification.
  3. [Theorem 1, Lemma 4] The MLCL bound is stated with empirical risks hat R_{D_i}(f^{t,i}) on the full previous datasets D_i, but in replay-based continual learning only memory subsets M_i are available at task t. The proof of Lemma 4 introduces ñ_i points per previous task without defining the corresponding empirical risk on memory or a memory-to-full-data gap term. Moreover, none of the terms in Theorem 1 depends on the particular memory-updating rule; the bound would hold for any replay strategy that satisfies the stated sample-size and imbalance quantities. Thus the claimed theoretical support for WRU specifically is absent.
  4. [Section E.2 and Appendix F] The limitations stated by the authors themselves apply directly to the central claim: Appendix F notes that finding optimal margins for all tasks remains challenging and that analyzing the actual deep networks via NTK is left to future work. In addition, the margin scale λ is a per-dataset fitted hyperparameter (λ = 1.0, 3.5, 4 for C-VOC, C-COCO, C-NUS), and all reported results come from a single run without error bars. Therefore the 'optimal margin' formula is not a falsifiable prediction, and the reported empirical superiority over baselines is not quantified against run-to-run variability.
minor comments (5)
  1. [Section 4, footnote 4] The footnote states that the hinge and logistic losses are 1-Lipschitz, which directly contradicts Assumption 2(3); the manuscript should either correct the assumption or remove the contradictory statement.
  2. [Section 3.1, footnote 3] The text 'prove that minimizing the RLADM loss' contains a typo: it should read RLDAM.
  3. [Section 6.2, Table 1] The caption says boldface denotes the best results, but only Macro-AUC rows are bolded; the Forgetting rows should be handled consistently, and the discussion should clarify whether lower forgetting is intended to be a strength of the method.
  4. [Section 6.4, Table 3] The ER results on C-COCO are non-monotonic in memory size (e.g., 62.68 at size 500, 63.97 at size 1000, 62.53 at size 1500), so the claim that the method is 'less sensitive to changes in memory size' should be supported by repeated runs and ideally confidence intervals.
  5. [Algorithm 1] The input list uses T for both the set of tasks and the task length, and the for loop over t ∈ (1, T) is ambiguous; distinct symbols would improve readability.

Circularity Check

1 steps flagged · score 6.0 of 10

The optimal-margin formula and the claimed margin-aware superiority reduce to the assumed ρ=1/Δ relation in Assumption 2(3); the margin law is a restatement of the assumption rather than a derivation.

  1. self definitional [Section 4, Assumption 2(3) and 'Choice of Optimal Margins'; Appendix B.3, Proposition 1]
    "ℓ(z − ∆+ k ) and ℓ(z − ∆− k ) is ρ+ k - and ρ− k -Lipschitz continuous with ρ+ k = 1/∆+ k , ρ− k = 1/∆− k . ... As ρ+ k = 1/∆+ k , we can adjust the margins ∆+ k and ∆+ k to make ρ+ k + ρ− k smaller under the constant constraint of ∆+ k + ∆− k , leading to better guarantees."

    The only place the margins enter the generalization bound is through the ρ± terms in Theorem 2, and those terms are fixed by Assumption 2(3) to be 1/Δ±. Proposition 1 then minimizes this bound over Δ+ + Δ− = β and 'derives' Δ*_k = λ/|D+_k|^{1/4}. With ρ = 1/Δ, that n^{-1/4} form is forced algebraically by the assumed inverse relation; the derivation returns exactly what the assumption encoded. The claimed superiority of RLDAM over RU, 'we can adjust the margins ... to make ρ+ k + ρ− k smaller ... leading to better guarantees', is likewise a restatement of the assumption: the bound improves only because the paper has defined the Lipschitz constants to shrink with larger margins.

full rationale

The paper's empirical contribution is not circular: RLDAM+WRU is evaluated against external baselines (ER, ER-RS, PRS, OCDM, KRT) on three benchmarks, and the reported gains are independent of the theory. The WRU memory-updating mechanism is also defined directly from the goal of preserving |D+_k|, |D−_k| ratios, and its effect on the continual-learning bound is a construction rather than a hidden reuse of the conclusion. The heavy borrowing from Wu, Li, and Yin (2023) is self-citation in the sense that two authors overlap, but the cited base theorem and Rademacher complexity lemmas are published, externally checkable results, so I do not count that as circularity by itself. The real circular step is in the theoretical derivation chain: Assumption 2(3) sets ρ±_k = 1/Δ±_k for the hinge loss. That inverse-margin relation is then the sole source of margin dependence in Theorem 2, and Proposition 1 minimizes that same bound to obtain Δ*_k = λ/|D+_k|^{1/4}. Thus the 'derived' optimal-margin law and the 'superior guarantee' of RLDAM over RU are algebraic consequences of an assumption that already contains the answer, rather than consequences of the loss, the Macro-AUC risk, or the hypothesis space. The paper even concedes in Appendix F that 'there remain challenges to find the optimal margins for all tasks', and in the experiments λ is tuned per dataset (1.0, 3.5, 4), so the margin scale is a fitted hyper-parameter, not a prediction. Because the central theoretical claim partially reduces to its own assumption, but the method and experiments stand independently, the appropriate circularity score is 6.

Assumptions & free parameters 1 free parameters · 4 assumptions · 0 invented entities

The central derivation depends on a questionable Lipschitz assumption and a fitted margin scale. The theory uses the authors' prior RU-loss framework; the MLCL portion follows Shi and Wang (2023) with Rademacher complexity substituted. No entities beyond the method itself are introduced.

free parameters (1)
  • lambda (margin scale) = 1.0 (C-VOC), 3.5 (C-COCO), 4 (C-NUS-WIDE); Fig 4 suggests 0.7 on C-VOC
    Tuned per dataset to maximize Macro-AUC; margins Delta+/- = lambda / |D+_k|^(1/4) depend on it.
assumptions (4)
  • domain assumption Assumption 1: tasks have disjoint label sets, each task has at least one multi-label sample, and samples may repeat across tasks
    Defines the MLCIL setting; the repeat-samples clause is used to build the benchmarks (Appendix E.1) and makes replay more favorable than it would be with disjoint samples.
  • ad hoc to paper Assumption 2(3): shifted hinge loss is (1/Delta)-Lipschitz
    Not true for the standard hinge loss, which is 1-Lipschitz; the entire margin-aware bound and optimal-margin derivation depend on this relation.
  • domain assumption Kernel-based hypothesis space (Eq. 7) with extension to deep ResNets via NTK
    The theory is proved for kernel models; experiments use ResNet-34/50, justified only by an NTK approximation that is not formally applied.
  • ad hoc to paper Two-label optimal-margin result extends to multi-label and to one-label tasks with equal positive/negative margins
    The extension from two labels to general multi-label tasks, and the binary equal-margin conclusion, are heuristic; the final margin schedule is set by hand rather than derived from the data.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Towards Macro-AUC oriented Imbalanced Multi-Label Continual Learning." pith.science (2026). https://pith.science/paper/UYICLL3S

@misc{pith2026241218231,
  author       = {Pith},
  title        = {Pith review of: Towards Macro-AUC oriented Imbalanced Multi-Label Continual Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UYICLL3S}},
  note         = {Machine review of arXiv:2412.18231}
}
read the original abstract

In Continual Learning (CL), while existing work primarily focuses on the multi-class classification task, there has been limited research on Multi-Label Learning (MLL). In practice, MLL datasets are often class-imbalanced, making it inherently challenging, a problem that is even more acute in CL. Due to its sensitivity to imbalance, Macro-AUC is an appropriate and widely used measure in MLL. However, there is no research to optimize Macro-AUC in MLCL specifically. To fill this gap, in this paper, we propose a new memory replay-based method to tackle the imbalance issue for Macro-AUC-oriented MLCL. Specifically, inspired by recent theory work, we propose a new Reweighted Label-Distribution-Aware Margin (RLDAM) loss. Furthermore, to be compatible with the RLDAM loss, a new memory-updating strategy named Weight Retain Updating (WRU) is proposed to maintain the numbers of positive and negative instances of the original dataset in memory. Theoretically, we provide superior generalization analyses of the RLDAM-based algorithm in terms of Macro-AUC, separately in batch MLL and MLCL settings. This is the first work to offer theoretical generalization analyses in MLCL to our knowledge. Finally, a series of experimental results illustrate the effectiveness of our method over several baselines. Our codes are available at https://github.com/ML-Group-SDU/Macro-AUC-CL.

Figures

Figures reproduced from arXiv: 2412.18231 by the authors.

Figure 1
Figure 1. The comparison of training curves between our [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. The comparison of overall test performances (test on [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Imbalance statistics of samples of each class in three tasks in C-MSCOCO. [PITH_FULL_IMAGE:figures/full_fig_p018_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: The effect of hyper-parameter λ to our RLDAM loss on C-PASCAL-VOC. F Ethical Statement Although we proposed to assign more proper margins in RLDAM loss to obtain a tighter generalization bound for both batch MLL and MLCL, there remain challenges to find the optimal mar…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

47 extracted references · 26 canonical work pages

  1. [1]

    , " * write output.state after.block = add.period write newline

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.a...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

  3. [3]

    Cao, K.; Wei, C.; Gaidon, A.; Arechiga, N.; and Ma, T. 2019. Learning imbalanced datasets with label-distribution-aware margin loss. Advances in neural information processing systems, 32

    2019

  4. [4]

    D.; Fabris, A.; and Susto, G

    Ceccon, M.; Pezze, D. D.; Fabris, A.; and Susto, G. A. 2024. Multi-Label Continual Learning for the Medical Domain: A Novel Benchmark. arXiv preprint arXiv:2404.06859

    arXiv 2024

  5. [5]

    K.; Ajanthan, T.; and Torr, P

    Chaudhry, A.; Dokania, P. K.; Ajanthan, T.; and Torr, P. H. 2018. Riemannian walk for incremental learning: Understanding forgetting and intransigence. In Proceedings of the European conference on computer vision (ECCV), 532--547

    2018

  6. [6]

    Chaudhry, A.; Rohrbach, M.; Elhoseiny, M.; Ajanthan, T.; Dokania, P.; Torr, P.; and Ranzato, M. 2019. Continual learning with tiny episodic memories. In Workshop on Multi-Task and Lifelong Reinforcement Learning

    work page 2019

  7. [7]

    Chua, T.-S.; Tang, J.; Hong, R.; Li, H.; Luo, Z.; and Zheng, Y. 2009. Nus-wide: a real-world web image database from national university of singapore. In Proceedings of the ACM international conference on image and video retrieval, 1--9

    work page 2009

  8. [8]

    Dalle Pezze, D.; Deronjic, D.; Masiero, C.; Tosato, D.; Beghi, A.; and Susto, G. A. 2023. A multi-label continual learning framework to scale deep learning approaches for packaging equipment monitoring. Engineering Applications of Artificial Intelligence, 124: 106610

    work page 2023

Show all 47 references
  1. [9]

    De Lange, M.; Aljundi, R.; Masana, M.; Parisot, S.; Jia, X.; Leonardis, A.; Slabaugh, G.; and Tuytelaars, T. 2021. A continual learning survey: Defying forgetting in classification tasks. IEEE transactions on pattern analysis and machine intelligence, 44(7): 3366--3385

    2021

  2. [10]

    Deng, J.; Dong, W.; Socher, R.; Li, L.-J.; Li, K.; and Fei-Fei, L. 2009. Imagenet: A large-scale hierarchical image database. In 2009 IEEE conference on computer vision and pattern recognition, 248--255. Ieee

    2009

  3. [11]

    Dong, S.; Luo, H.; He, Y.; Wei, X.; Cheng, J.; and Gong, Y. 2023. Knowledge Restore and Transfer for Multi-Label Class-Incremental Learning. In Proceedings of the IEEE/CVF International Conference on Computer Vision, 18711--18720

    2023

  4. [12]

    A.; Van Gool, L.; Williams, C

    Everingham, M.; Eslami, S. A.; Van Gool, L.; Williams, C. K.; Winn, J.; and Zisserman, A. 2015. The pascal visual object classes challenge: A retrospective. International journal of computer vision, 111: 98--136

    2015

  5. [13]

    Evron, I.; Moroshko, E.; Buzaglo, G.; Khriesh, M.; Marjieh, B.; Srebro, N.; and Soudry, D. 2023. Continual Learning in Linear Classification on Separable Data. arXiv preprint arXiv:2306.03534

    2023 arXiv

  6. [14]

    Farajtabar, M.; Azizan, N.; Mott, A.; and Li, A. 2020. Orthogonal gradient descent for continual learning. In International Conference on Artificial Intelligence and Statistics, 3762--3773. PMLR

    2020

  7. [15]

    A.; Pritzel, A.; and Wierstra, D

    Fernando, C.; Banarse, D.; Blundell, C.; Zwols, Y.; Ha, D.; Rusu, A. A.; Pritzel, A.; and Wierstra, D. 2017. Pathnet: Evolution channels gradient descent in super neural networks. arXiv preprint arXiv:1701.08734

    2017 arXiv

  8. [16]

    French, R. M. 1999. Catastrophic forgetting in connectionist networks. Trends in cognitive sciences, 3(4): 128--135

    1999

  9. [17]

    He, K.; Zhang, X.; Ren, S.; and Sun, J. 2016. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, 770--778

    2016

  10. [18]

    Huang, K.; Wang, Y.; Tao, M.; and Zhao, T. 2020. Why Do Deep Residual Networks Generalize Better than Deep Feedforward Networks?---A Neural Tangent Kernel Perspective. Advances in neural information processing systems, 33: 2698--2709

    2020

  11. [19]

    Jacot, A.; Gabriel, F.; and Hongler, C. 2018. Neural tangent kernel: Convergence and generalization in neural networks. Advances in neural information processing systems, 31

    2018

  12. [20]

    Kassim, M. A. 2024. Multi-label Lifelong Machine Learning using Deep Generative Replay. Ph.D. thesis, Universit \'e d'Ottawa| University of Ottawa

    2024

  13. [21]

    D.; Jeong, J.; and Kim, G

    Kim, C. D.; Jeong, J.; and Kim, G. 2020. Imbalanced continual learning with partitioning reservoir sampling. In Computer Vision--ECCV 2020: 16th European Conference, Glasgow, UK, August 23--28, 2020, Proceedings, Part XIII 16, 411--428. Springer

    2020

  14. [22]

    A.; Milan, K.; Quan, J.; Ramalho, T.; Grabska-Barwinska, A.; et al

    Kirkpatrick, J.; Pascanu, R.; Rabinowitz, N.; Veness, J.; Desjardins, G.; Rusu, A. A.; Milan, K.; Quan, J.; Ramalho, T.; Grabska-Barwinska, A.; et al. 2017. Overcoming catastrophic forgetting in neural networks. Proceedings of the national academy of sciences, 114(13): 3521--3526

    2017

  15. [23]

    Knoblauch, J.; Husain, H.; and Diethe, T. 2020 a . Optimal continual learning has perfect memory and is np-hard. In International Conference on Machine Learning, 5327--5337. PMLR

    2020

  16. [24]

    Knoblauch, J.; Husain, H.; and Diethe, T. 2020 b . Optimal continual learning has perfect memory and is np-hard. In International Conference on Machine Learning, 5327--5337. PMLR

    2020

  17. [25]

    Liang, Y.-S.; and Li, W.-J. 2022. Optimizing Class Distribution in Memory for Multi-Label Online Continual Learning. arXiv preprint arXiv:2209.11469

    2022 arXiv

  18. [26]

    Lin, S.; Ju, P.; Liang, Y.; and Shroff, N. 2023. Theory on Forgetting and Generalization of Continual Learning. arXiv preprint arXiv:2302.05836

    2023 arXiv

  19. [27]

    Lin, T.-Y.; Maire, M.; Belongie, S.; Hays, J.; Perona, P.; Ramanan, D.; Doll \'a r, P.; and Zitnick, C. L. 2014. Microsoft coco: Common objects in context. In Computer Vision--ECCV 2014: 13th European Conference, Zurich, Switzerland, September 6-12, 2014, Proceedings, Part V 1...

    2014

  20. [28]

    L.; De Lange, M.; Masana, M.; Pomponi, J.; Van de Ven, G

    Lomonaco, V.; Pellegrini, L.; Cossu, A.; Carta, A.; Graffieti, G.; Hayes, T. L.; De Lange, M.; Masana, M.; Pomponi, J.; Van de Ven, G. M.; et al. 2021. Avalanche: an end-to-end library for continual learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pat...

    2021

  21. [29]

    Mallya, A.; and Lazebnik, S. 2018. Packnet: Adding multiple tasks to a single network by iterative pruning. In Proceedings of the IEEE conference on Computer Vision and Pattern Recognition, 7765--7773

    2018

  22. [30]

    Mansour, Y.; Mohri, M.; and Rostamizadeh, A. 2009. Domain adaptation: Learning bounds and algorithms. arXiv preprint arXiv:0902.3430

    2009 arXiv

  23. [31]

    McCallum, A. K. 1999. Multi-label text classification with a mixture model trained by EM . In AAAI 99 workshop on Text Learning. Citeseer

    1999

  24. [32]

    Mohri, M.; Rostamizadeh, A.; and Talwalkar, A. 2018. Foundations of machine learning. MIT press

    2018

  25. [33]

    Rebuffi, S.-A.; Kolesnikov, A.; Sperl, G.; and Lampert, C. H. 2017. icarl: Incremental classifier and representation learning. In Proceedings of the IEEE conference on Computer Vision and Pattern Recognition, 2001--2010

    2017

  26. [34]

    Riemer, M.; Cases, I.; Ajemian, R.; Liu, M.; Rish, I.; Tu, Y.; and Tesauro, G. 2018. Learning to learn without forgetting by maximizing transfer and minimizing interference. arXiv preprint arXiv:1810.11910

    2018 arXiv

  27. [35]

    Ring, M. B. 1994. Continual learning in reinforcement environments. The University of Texas at Austin

    1994

  28. [36]

    Robbins, H.; and Monro, S. 1951. A stochastic approximation method. The annals of mathematical statistics, 400--407

    1951

  29. [37]

    Shi, H.; and Wang, H. 2023. A Unified Approach to Domain Incremental Learning with Memory: Theory and Algorithm. Advances in Neural Information Processing Systems, 36

    2023

  30. [38]

    Shmelkov, K.; Schmid, C.; and Alahari, K. 2017. Incremental learning of object detectors without catastrophic forgetting. In Proceedings of the IEEE international conference on computer vision, 3400--3409

    2017

  31. [39]

    N.; Giacobini, M.; and Michalak, K

    Tarekegn, A. N.; Giacobini, M.; and Michalak, K. 2021. A review of methods for imbalanced multi-label classification. Pattern Recognition, 118: 107965

    2021

  32. [40]

    Tirer, T.; Bruna, J.; and Giryes, R. 2022. Kernel-based smoothness analysis of residual networks. In Mathematical and Scientific Machine Learning, 921--954. PMLR

    2022

  33. [41]

    Tsoumakas, G.; and Katakis, I. 2007. Multi-label classification: An overview. International Journal of Data Warehousing and Mining (IJDWM), 3(3): 1--13

    2007

  34. [42]

    Vitter, J. S. 1985. Random sampling with a reservoir. ACM Transactions on Mathematical Software (TOMS), 11(1): 37--57

    1985

  35. [43]

    Wang, L.; Zhang, X.; Su, H.; and Zhu, J. 2023. A comprehensive survey of continual learning: Theory, method and application. arXiv preprint arXiv:2302.00487

    2023 arXiv

  36. [44]

    Wu, G.; Li, C.; and Yin, Y. 2023. Towards Understanding Generalization of Macro-AUC in Multi-label Learning. International Conference on Machine Learning

    2023

  37. [45]

    Zenke, F.; Poole, B.; and Ganguli, S. 2017. Continual learning through synaptic intelligence. In International conference on machine learning, 3987--3995. PMLR

    2017

  38. [46]

    Zhang, M.-L.; and Zhou, Z.-H. 2013 a . A review on multi-label learning algorithms. IEEE transactions on knowledge and data engineering, 26(8): 1819--1837

    2013

  39. [47]

    Zhang, M.-L.; and Zhou, Z.-H. 2013 b . A review on multi-label learning algorithms. IEEE transactions on knowledge and data engineering, 26(8): 1819--1837

    2013

Pith tools

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