Pith. sign in

REVIEW 4 major objections 4 minor 47 references

Make Domain Shift a Catastrophic Forgetting Alleviator in Class-Incremental Learning

T0 review · 4 major / 4 minor · reviewed 2026-08-10 · deepseek-v4-flash

Pith's one-line read Domain shift cuts forgetting in class-incremental learning

desk verdict The headline observation about domain shift reducing forgetting is likely a task-boundary artifact, but the DisCo method stands on solid unconfounded benchmark gains. read the letter →

arxiv 2501.00237 v1 pith:LDGRCOCT submitted 2024-12-31 cs.CV cs.LG

classification cs.CVcs.LG
keywords class-incrementallearningcatastrophicforgettingdomainshiftcontrastiveprototypepoolfeatureseparationcontinualrehearsal
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 a per-task domain shift makes class-incremental learning forget less, not more. In its CILD scenario—CIFAR-100 restyled per task with a style-transfer model, or DomainNet split across its six domains—five standard methods (iCaRL, BiC, MEMO, LwF, DER) all show substantially lower forgetting than in the same-class CIL protocol, while prompt-based L2P is the exception. The explanation proposed is that domain shifts separate task feature distributions and thereby reduce parameter interference between successive tasks. Building on that, the paper introduces DisCo, a rehearsal-based add-on that uses a lightweight prototype pool with contrastive losses to push current-task features away from previous-task prototypes and distill old-class features from a frozen teacher. If the interpretation is right, separating features across tasks is itself a general forgetting countermeasure, independent of any real style change.

What carries the argument

The central mechanism is the task-prototype pool with contrastive regularization. For each task, a momentum-averaged prototype is stored—either from batch image features or from CLIP text embeddings of the class names; a task-level triplet loss pulls current samples toward the current prototype and pushes them away from all prior prototypes, while a class-level triplet loss pulls samples of the same class together within the current task. A cross-task contrastive distillation loss then aligns the current student's features for replayed old samples with the frozen teacher's features of the same class. Together these losses create a margin between tasks in feature space, which is the paper's proposed stand-in for the input-level domain separation observed in the empirical study.

What would settle it

Evaluate a CILD-trained model on test samples whose domain cue is scrambled—e.g., test every task's classes under a style it never saw, or force a single shared classifier that has no way to select a per-task head. If the forgetting measure returns to CIL levels, the central claim collapses into task-incremental advantages; if it stays low, the separation itself is what protects old knowledge.

Watch

Extended reading notes

Core claim

The paper's discovery claim is that introducing domain shift into the input distribution of each incremental task reduces catastrophic forgetting across a broad set of CIL methods. On DomainCIFAR-100 and DomainNet, the forgetting measure FM drops sharply under the CILD protocol compared with standard CIL—for example DER falls to 0.50 from 40.26 on CIFAR-100—and t-SNE shows task clusters that are more separated. The paper ascribes the drop to lower parameter interference, quantified by its PIV and PFTS metrics: when the input statistics of a task are distinctive, the high-magnitude parameter updates for different tasks have less overlap. Based on this observation, DisCo is put forward as a method-level transfer of the benefit: instead of changing the input domain, it imposes task-level and class-level contrastive regularization around a prototype pool and adds cross-task contrastive distillation, so that ordinary CIL tasks acquire separated feature distributions reminiscent of domain-shifted inputs.

Load-bearing premise

The load-bearing premise is that the low forgetting measured under the domain-shift protocol (CILD) reflects genuine retention of old knowledge, not exploitation of the per-task domain style as a task-identity cue at inference.

Editorial extensions

If this is right

  • If the observation is correct, a gradual style or domain change over the task stream is not a hazard to be corrected but a protective signal that lowers interference.
  • Plugging DisCo into rehearsal-based baselines reduces forgetting reliably across CIFAR-100, Fashion-MNIST, and Tiny-ImageNet; on CIFAR-100 iCaRL's FM drops from 51.34 to 33.96 and its AA rises from 64.24 to 70.11.
  • Task-level regularization is the main driver of the forgetting reduction; class-level regularization raises initial task accuracy; cross-task distillation gives an additional retention boost.
  • DisCo generalizes in plug-in form to regularization-based and prompt-based methods, with text prototypes working better for prompt methods than image prototypes.
  • A corollary: separating task features is sufficient for a large part of the anti-forgetting benefit, so methods need not transform the input domain to profit from domain-shift-like separation.

Reading between the lines

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

  • Editorial inference: the paper's own task-inference-accuracy numbers show CILD models can detect the task through the domain cue, so the cleanest test of the causal story is to remove that cue at test time; a shared-classifier variant would separate genuine retention from task-boundary recall.
  • Editorial extension: a testable prediction of the parameter-interference story is that task-specific random input perturbations (e.g., noise or color shifts uncorrelated with class semantics) would reproduce part of the forgetting reduction; if they do, input diversification, not the specific domain content, is the protective factor.
  • Editorial connection: the prototype-pool mechanism is modality-agnostic, so applying the same task-level margin to audio or text continual learning is a natural next check; the paper does not claim this, but nothing in the method depends on images.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 4 minor

Summary. The paper studies whether introducing a per-task domain shift into class-incremental learning (CILD) reduces catastrophic forgetting. It reports that five of six baselines show lower forgetting under CILD than under standard CIL on CIFAR-100 and DomainNet, and attributes this to clearer feature separation and reduced parameter interference. Based on this observation, the authors propose DisCo, a plug-and-play contrastive regularization method that uses a prototype pool and cross-task distillation to keep current-task features away from previous-task prototypes. DisCo is evaluated on standard CIL benchmarks (CIFAR-100, Fashion-MNIST, Tiny-ImageNet, CUB200) and reported to improve average accuracy and reduce forgetting when added to iCaRL, BiC, LwF, DER, and L2P.

Significance. The observation, if valid, would be surprising and practically useful: it would suggest that input-level domain diversity can protect old class knowledge in CIL and that feature-space separation is a general forgetting countermeasure. The paper ships code, covers multiple backbones, pretraining regimes, domain orders, and datasets, and includes component ablations. However, the primary evidence is not CIL-valid: the CILD protocol leaks task identity through domain-aligned test sets, and the paper's own Table 8 and Section 4 confirm this concern. The DisCo results are independent but modest and lack statistical reporting.

major comments (4)
  1. [Section 3.2, Table 1, Appendix B.3 (Table 8)] The central observation that CILD reduces forgetting is confounded by task-boundary detection. In CILD each task's test set is drawn from a unique domain, so at inference the model can infer the task identity from the input style and exploit a task-specific decision path. Table 8 shows exactly this: TIA for iCaRL rises from 50.14 under CIL to 84.29 under CILD. The rebuttal using ITA (79.06 vs 36.22) manually routes each sample to its task-specific classifier head, which is a task-incremental evaluation protocol, not the class-incremental protocol the paper claims to defend. Because FM (Eq. 10-11) is computed on domain-aligned test sets, the lower FM in Table 1 may reflect task-boundary exploitation rather than reduced forgetting of generalizable class knowledge. The paper needs a task-agnostic evaluation (e.g., testing all tasks on a shared domain) to separate these explanations.
  2. [Section 4, Appendix D] The authors themselves state in Section 4 that applying domain shift at the input level requires the task ID at inference and "contradicts the CIL setting." Appendix D repeats that at inference one needs to decide which task a sample belongs to. This admission directly undermines the transfer from the CILD observation to a CIL claim. The proposed DisCo method simulates domain shift at the feature level, but the paper does not demonstrate that feature separation is equivalent to input-level domain shift or that the CILD numbers are a valid proxy for the benefit of DisCo in CIL.
  3. [Table 2, Appendix C.5] No standard deviations or per-run values are reported for any of the 3-run averages in Table 2. Several improvements are small relative to plausible run-to-run variation (e.g., L2P + DisCo-I AA 82.78 vs 82.65 and FM 7.98 vs 7.62; DER + DisCo-I on Tiny-ImageNet AA 12.08 vs 11.58). The paper's claim of "substantial performance improvements" is consequently not statistically supported.
  4. [Appendix A.2, Eqs. (13)-(16)] The PIV and PFTS metrics are introduced without validation. They measure the Jaccard overlap of high-magnitude parameter updates, but no experiment links this overlap to actual forgetting or accuracy retention. Under DisCo's task-level contrastive loss (Eq. 3), the current task's features are explicitly pushed away from previous prototypes, so a reduction in interference is expected by construction; the paper does not show that PIV/PFTS are causally related to the forgetting measure FM.
minor comments (4)
  1. [Section 5.2] The sentence "increase AA by 3.31% and reduce FM by 10.94% on Tiny-ImageNet" does not match Table 2 for iCaRL, where the changes are +2.30 AA and -11.21 FM; please correct the numbers or clarify which comparison is meant.
  2. [Appendix B.3, Table 8] The column heading "IT A − f irst" is malformed; it should read "ITA (first task)" or similar.
  3. [Section 4, Eq. (2)] The momentum update in Eq. (2) uses p_i on both sides with the same symbol; please use a distinct notation for the batch prototype and the previous running estimate, and state the initialization.
  4. [Throughout] There are several typographical issues: "dis-joint" should be "disjoint", "P IVand" needs spacing, "T riplet" appears in equations instead of "Triplet", and "V olpi" in the references should be "Volpi".

Circularity Check

2 steps flagged · score 6.0 of 10

CILD's lower FM is protocol-enforced via domain-as-task-ID test sets; DisCo's external benchmark gains remain independent.

  1. self definitional [Section 3.2 (CILD definition, Evaluation Protocols), Table 1, Appendix A.1 Eq. (10)-(11), Section 4]
    "CILD: Based on CIL, each task t is modified by introducing a unique variation in the domain while sharing the same label space Ct with CIL. ... fj,k = max_{i∈{1,...,k−1}}(ai,j − ak,j), ∀j < k. FM = 1/(T−1) Σ_{j=1}^{T−1} fj,T−1. ... This resembles the paradigm of task-incremental learning (Wang et al. 2024), where the task ID (indicating the domain shift in our case) is needed at inference, which contradicts the CIL setting."

    Under CILD, every test sample of task j comes from task j's unique domain, so the input style itself reveals the task identity (the paper's own Table 8 gives iCaRL TIA 50.14 under CIL vs 84.29 under CILD). The FM definition in Eq. (10)-(11) only compares each task's peak and final accuracy on that same domain-aligned test set. A model that learns to use domain as a task cue can therefore keep FM low without retaining generalizable class knowledge. The authors explicitly admit that applying domain shift at input level resembles task-incremental learning and 'contradicts the CIL setting.' Thus the Table 1 observation that domain shift reduces forgetting is forced by the scenario construction plus the FM metric; it is the known task-incremental advantage re-encoded in the CILD protocol.

  2. other [Appendix B.3, 'Dive into the forgetting of CILD']
    "For ITA, we isolate the weight of the classifier and treat each of them as the exclusive classifier for each task. We manually feed the sample feature to its corresponding classifier and get its ITA in that classifier. We only compare the ITA of the first task since they share the same test set."

    The paper uses Intra-Task Accuracy to rebut the task-boundary confound, claiming 'the model does forget what it learned.' But ITA manually routes every test feature to the classifier head of the corresponding task, which is exactly a task-incremental evaluation protocol, not the class-incremental setting the paper claims to defend. This removes the task-identification burden that defines CIL. The high ITA under CILD therefore shows only that task-specific heads retain class knowledge when the task ID is supplied by the experimenter; it does not independently establish reduced catastrophic forgetting in CIL. The rebuttal re-imports the task identity that the CILD protocol already provides.

full rationale

The paper has two distinct claims: (1) the empirical observation that domain shift reduces forgetting in CIL, and (2) the DisCo method that boosts existing CIL methods. The first claim is substantially confounded by construction: CILD gives each task a unique domain, and FM is measured on those same domain-aligned test sets, so the input style leaks the task ID. The paper's own TIA table and its admission that input-level domain shift 'resembles the paradigm of task-incremental learning' confirm this. The ITA rebuttal uses manual routing to task-specific classifiers, which is a task-incremental protocol and cannot rescue the CIL-specific claim. I therefore score the observation as partially circular: the lower FM under CILD reduces, to a significant degree, to the protocol encoding task identity. The second claim, DisCo, is evaluated on standard CIL benchmarks without domain shift (Table 2), and those AA/FM numbers are independent of the loss definitions; the task-level loss makes feature separation true by construction, but the reported accuracy gains are external and genuine. There is no load-bearing self-citation chain: the self-citations (Luo et al. 2024; Lai et al. 2024; Liu and Zhou 2024) are background references and do not carry the derivation. Overall, the central empirical motivation is protocol-enforced, while the method's external evaluation remains independent, giving a partial-circularity score of 6.

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

The ledger reflects what DisCo and the observation pull from outside the paper. The three loss weights are tuned on CIFAR-100 using the appendix grid, a mild form of fitted parameters, and the PIV threshold is hand-chosen, so they are counted as free parameters. The domain assumptions carry the interpretive load: domain-aligned test sets, style-transfer content preservation, faithful PILOT baseline implementations, and PIV as a forgetting proxy. The paper introduces one algorithmic entity (the prototype pool) and one dataset artifact (DomainCIFAR-100), plus two diagnostic metrics, each with falsifiable handles in benchmark results or reproducible code, so the graviton problem does not apply.

free parameters (4)
  • lambda_tcon = 0.5
    Task-level contrastive loss weight (Eq. 7), chosen by hand from the Appendix C.5 grid on CIFAR-100; reused across all benchmarks.
  • lambda_ccon = 0.5
    Class-level contrastive loss weight, tuned jointly with lambda_tcon on CIFAR-100; higher values generally lower AA (Table 10).
  • lambda_ccd = 1.0
    Cross-task contrastive distillation weight (Eq. 7), tuned on CIFAR-100; lowering it to 0.5 worsens FM for iCaRL and BiC.
  • PIV threshold delta = upper quartile of per-task parameter-update magnitudes
    Defines the high-magnitude update set in PIV (Appendix A.2); the metric's values and the interference conclusion depend on this hand-chosen cutoff.
assumptions (4)
  • domain assumption CILD evaluation assumes domain-aligned test sets, making domain identity a stable inference-time cue for old classes.
    Underlies the Table 1 FM comparison; the paper's TIA analysis (Table 8) confirms the cue is used, so the measured forgetting reduction mixes task-boundary exploitation with genuine retention.
  • domain assumption AvatarNet style transfer preserves the class-discriminative content needed for DomainCIFAR-100 to be a valid benchmark.
    Appendix B.1 constructs the CILD benchmark by style transfer; if style synthesis corrupts labels, the observation would be an artifact of the synthesis pipeline rather than of domain shift.
  • domain assumption Baseline implementations from LAMDA-PILOT faithfully match the original published methods.
    All comparisons in Tables 1 and 2 run PILOT versions of iCaRL, BiC, LwF, DER, L2P, and MEMO; implementation drift could shift the absolute numbers and inflate DisCo's relative gains.
  • ad hoc to paper PIV and PFTS, computed from parameter-update overlap, are valid proxies for forgetting and plasticity.
    Introduced in Appendix A.2 with an arbitrary upper-quartile threshold; the mapping from update-set overlap to final test accuracy is assumed, not derived or causally tested.
invented entities (3)
  • DisCo prototype pool (momentum-accumulated per-task prototypes) independent evidence
    purpose: Anchor vectors for task-level and class-level contrastive losses; simulates domain-shift separation in feature space rather than input space.
    Its effect is falsifiable on standard CIL benchmarks (Table 2) and through the public repository; algorithmic component, not a physical postulate.
  • DomainCIFAR-100 independent evidence
    purpose: Synthesized CILD benchmark adding five AvatarNet styles to CIFAR-100 for controlled domain-shift experiments.
    Reproducible from public AvatarNet and CIFAR-100 artifacts; validity depends on the style-transfer content-preservation assumption.
  • PIV and PFTS metrics independent evidence
    purpose: Quantify cross-task parameter interference and forward transfer to support the mechanism claim.
    Computable from checkpoints of any run; validity as causal probes for forgetting is not independently established outside this paper.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Make Domain Shift a Catastrophic Forgetting Alleviator in Class-Incremental Learning." pith.science (2026). https://pith.science/paper/LDGRCOCT

@misc{pith2026250100237,
  author       = {Pith},
  title        = {Pith review of: Make Domain Shift a Catastrophic Forgetting Alleviator in Class-Incremental Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LDGRCOCT}},
  note         = {Machine review of arXiv:2501.00237}
}
read the original abstract

In the realm of class-incremental learning (CIL), alleviating the catastrophic forgetting problem is a pivotal challenge. This paper discovers a counter-intuitive observation: by incorporating domain shift into CIL tasks, the forgetting rate is significantly reduced. Our comprehensive studies demonstrate that incorporating domain shift leads to a clearer separation in the feature distribution across tasks and helps reduce parameter interference during the learning process. Inspired by this observation, we propose a simple yet effective method named DisCo to deal with CIL tasks. DisCo introduces a lightweight prototype pool that utilizes contrastive learning to promote distinct feature distributions for the current task relative to previous ones, effectively mitigating interference across tasks. DisCo can be easily integrated into existing state-of-the-art class-incremental learning methods. Experimental results show that incorporating our method into various CIL methods achieves substantial performance improvements, validating the benefits of our approach in enhancing class-incremental learning by separating feature representation and reducing interference. These findings illustrate that DisCo can serve as a robust fashion for future research in class-incremental learning.

Figures

Figures reproduced from arXiv: 2501.00237 by the authors.

Figure 1
Figure 1. The key finding of our work: Incorporating domain [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Illustration of two scenarios construction on two datasets respectively. In Fig. 2(a), we use AvatarNet (Sheng et al. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. t-SNE visualization of features on CIFAR-100. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: The overview framework of DisCo. DisCo includes Task&Class -level Regularization and Cross-task Contrastive [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Ablation study on incremental task length. [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Comparison of accuracy curves of the first task on [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]
Figure 7
Figure 7. Figure 7: t-SNE visualization of features on CIFAR-100. The top row denotes features extracted by different baseline continual [PITH_FULL_IMAGE:figures/full_fig_p014_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

47 extracted references · 25 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]

    Buzzega, P.; Boschini, M.; Porrello, A.; Abati, D.; and Calderara, S. 2020. Dark experience for general continual learning: a strong, simple baseline. Advances in neural information processing systems, 33: 15920--15930

  4. [4]

    Caccia, L.; Belilovsky, E.; Caccia, M.; and Pineau, J. 2020. Online learned continual compression with adaptive quantization modules. In International conference on machine learning, 1240--1250. PMLR

  5. [5]

    Cha, H.; Lee, J.; and Shin, J. 2021. Co2l: Contrastive continual learning. In Proceedings of the IEEE/CVF International conference on computer vision, 9516--9525

  6. [6]

    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

  7. [7]

    Dosovitskiy, A.; Beyer, L.; Kolesnikov, A.; Weissenborn, D.; Zhai, X.; Unterthiner, T.; Dehghani, M.; Minderer, M.; Heigold, G.; Gelly, S.; et al. 2020. An image is worth 16x16 words: Transformers for image recognition at scale. arXiv preprint arXiv:2010.11929

  8. [8]

    Golkar, S.; Kagan, M.; and Cho, K. 2019. Continual learning via neural pruning. arXiv preprint arXiv:1903.04476

Show all 47 references
  1. [9]

    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

  2. [10]

    Jaccard, P. 1901. \'E tude comparative de la distribution florale dans une portion des Alpes et des Jura. Bull Soc Vaudoise Sci Nat, 37: 547--579

  3. [11]

    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

  4. [12]

    Krizhevsky, A.; and Hinton, G. 2009. Learning multiple layers of features from tiny images. Handbook of Systemic Autoimmune Diseases, 1(4)

  5. [13]

    N.; Venkatesh, R

    Kundu, J. N.; Venkatesh, R. M.; Venkat, N.; Revanur, A.; and Babu, R. V. 2020. Class-incremental domain adaptation. In Computer Vision--ECCV 2020: 16th European Conference, Glasgow, UK, August 23--28, 2020, Proceedings, Part XIII 16, 53--69. Springer

  6. [14]

    Lai, Y.; Zhou, Y.; Liu, X.; and Zhou, T. 2024. Memory-Assisted Sub-Prototype Mining for Universal Domain Adaptation. In The Twelfth International Conference on Learning Representations

  7. [15]

    Li, Z.; and Hoiem, D. 2017. Learning without forgetting. IEEE transactions on pattern analysis and machine intelligence, 40(12): 2935--2947

  8. [16]

    Lin, G.; Chu, H.; and Lai, H. 2022. Towards better plasticity-stability trade-off in incremental learning: A simple linear connector. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 89--98

  9. [17]

    D.; Jui, S.; and de Weijer, J

    Liu, X.; Wu, C.; Menta, M.; Herranz, L.; Raducanu, B.; Bagdanov, A. D.; Jui, S.; and de Weijer, J. v. 2020. Generative feature replay for class-incremental learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, 226--227

  10. [18]

    Liu, X.; and Zhou, Y. 2024. COCA: Classifier-Oriented Calibration via Textual Prototype for Source-Free Universal Domain Adaptation. In Proceedings of the Asian Conference on Computer Vision, 1671--1687

  11. [19]

    Lopez-Paz, D.; and Ranzato, M. 2017. Gradient episodic memory for continual learning. Advances in neural information processing systems, 30

  12. [20]

    Luo, S.; Chen, W.; Tian, W.; Liu, R.; Hou, L.; Zhang, X.; Shen, H.; Wu, R.; Geng, S.; Zhou, Y.; et al. 2024. Delving into Multi-modal Multi-task Foundation Models for Road Scene Understanding: From Learning Paradigm Perspectives. IEEE Transactions on Intelligent Vehicles

  13. [21]

    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

  14. [22]

    McCloskey, M.; and Cohen, N. J. 1989. Catastrophic interference in connectionist networks: The sequential learning problem. In Psychology of learning and motivation, volume 24, 109--165. Elsevier

  15. [23]

    Peng, X.; Bai, Q.; Xia, X.; Huang, Z.; Saenko, K.; and Wang, B. 2019. Moment matching for multi-source domain adaptation. In Proceedings of the IEEE/CVF international conference on computer vision, 1406--1415

  16. [24]

    Pham, Q.; Liu, C.; and Hoi, S. 2021. Dualnet: Continual learning, fast and slow. Advances in Neural Information Processing Systems, 34: 16131--16144

  17. [25]

    W.; Hallacy, C.; Ramesh, A.; Goh, G.; Agarwal, S.; Sastry, G.; Askell, A.; Mishkin, P.; Clark, J.; et al

    Radford, A.; Kim, J. W.; Hallacy, C.; Ramesh, A.; Goh, G.; Agarwal, S.; Sastry, G.; Askell, A.; Mishkin, P.; Clark, J.; et al. 2021. Learning transferable visual models from natural language supervision. In International conference on machine learning, 8748--8763. PMLR

  18. [26]

    Razdaibiedina, A.; Mao, Y.; Hou, R.; Khabsa, M.; Lewis, M.; and Almahairi, A. 2023. Progressive prompts: Continual learning for language models. arXiv preprint arXiv:2301.12314

  19. [27]

    Rebuffi, S.; Kolesnikov, A.; and Lampert, C. H. 2016. icarl: Incremental classifier and representation learning. CoRR abs/1611.07725 (2016). arXiv preprint arXiv:1611.07725

  20. [28]

    A.; Rabinowitz, N

    Rusu, A. A.; Rabinowitz, N. C.; Desjardins, G.; Soyer, H.; Kirkpatrick, J.; Kavukcuoglu, K.; Pascanu, R.; and Hadsell, R. 2016. Progressive neural networks. arXiv preprint arXiv:1606.04671

  21. [29]

    Sheng, L.; Lin, Z.; Shao, J.; and Wang, X. 2018. Avatar-net: Multi-scale zero-shot style transfer by feature decoration. In Proceedings of the IEEE conference on computer vision and pattern recognition, 8242--8250

  22. [30]

    Simon, C.; Faraki, M.; Tsai, Y.-H.; Yu, X.; Schulter, S.; Suh, Y.; Harandi, M.; and Chandraker, M. 2022. On generalizing beyond domains in cross-domain continual learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 9265--9274

  23. [31]

    S.; Karlinsky, L.; Gutta, V.; Cascante-Bonilla, P.; Kim, D.; Arbelle, A.; Panda, R.; Feris, R.; and Kira, Z

    Smith, J. S.; Karlinsky, L.; Gutta, V.; Cascante-Bonilla, P.; Kim, D.; Arbelle, A.; Panda, R.; Feris, R.; and Kira, Z. 2023. Coda-prompt: Continual decomposed attention-based prompting for rehearsal-free continual learning. In Proceedings of the IEEE/CVF Conference on Computer...

  24. [32]

    Sun, H.-L.; Zhou, D.-W.; Ye, H.-J.; and Zhan, D.-C. 2023. PILOT: A Pre-Trained Model-Based Continual Learning Toolbox. arXiv preprint arXiv:2309.07117

  25. [33]

    Tang, S.; Su, P.; Chen, D.; and Ouyang, W. 2021. Gradient regularized contrastive learning for continual domain adaptation. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 35, 2665--2673

  26. [34]

    Tao, X.; Hong, X.; Chang, X.; and Gong, Y. 2020. Bi-Objective Continual Learning: Learning ‘New’ While Consolidating ‘Known’. Proceedings of the AAAI Conference on Artificial Intelligence, 34(04): 5989--5996

  27. [35]

    M.; Siegelmann, H

    Van de Ven, G. M.; Siegelmann, H. T.; and Tolias, A. S. 2020. Brain-inspired replay for continual learning with artificial neural networks. Nature communications, 11(1): 4069

  28. [36]

    Van der Maaten, L.; and Hinton, G. 2008. Visualizing data using t-SNE. Journal of machine learning research, 9(11)

  29. [37]

    Volpi, R.; Larlus, D.; and Rogez, G. 2021. Continual adaptation of visual representations via domain randomization and meta-learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 4443--4453

  30. [38]

    Wah, C.; Branson, S.; Welinder, P.; Perona, P.; and Belongie, S. 2011. The caltech-ucsd birds-200-2011 dataset

  31. [39]

    Wang, L.; Zhang, X.; Su, H.; and Zhu, J. 2024. A comprehensive survey of continual learning: Theory, method and application. IEEE Transactions on Pattern Analysis and Machine Intelligence

  32. [40]

    Wang, Z.; Liu, L.; Duan, Y.; and Tao, D. 2022 a . Continual learning through retrieval and imagination. In Proceedings of the AAAI Conference on Artificial Intelligence, 8, 8594--8602

  33. [41]

    Wang, Z.; Zhang, Z.; Ebrahimi, S.; Sun, R.; Zhang, H.; Lee, C.-Y.; Ren, X.; Su, G.; Perot, V.; Dy, J.; et al. 2022 b . Dualprompt: Complementary prompting for rehearsal-free continual learning. In European Conference on Computer Vision, 631--648. Springer

  34. [42]

    Wang, Z.; Zhang, Z.; Lee, C.-Y.; Zhang, H.; Sun, R.; Ren, X.; Su, G.; Perot, V.; Dy, J.; and Pfister, T. 2022 c . Learning to prompt for continual learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 139--149

  35. [43]

    Wu, Y.; Chen, Y.; Wang, L.; Ye, Y.; Liu, Z.; Guo, Y.; and Fu, Y. 2019. Large scale incremental learning. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 374--382

  36. [44]

    Xiao, H.; Rasul, K.; and Vollgraf, R. 2017. Fashion-mnist: a novel image dataset for benchmarking machine learning algorithms. arXiv preprint arXiv:1708.07747

  37. [45]

    Xie, J.; Yan, S.; and He, X. 2022. General incremental learning with domain-aware categorical representations. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 14351--14360

  38. [46]

    Yao, L.; and Miller, J. 2015. Tiny imagenet classification with convolutional neural networks. CS 231N, 2(5): 8

  39. [47]

    Zhou, D.-W.; Wang, Q.-W.; Ye, H.-J.; and Zhan, D.-C. 2022. A model or 603 exemplars: Towards memory-efficient class-incremental learning. arXiv preprint arXiv:2205.13218

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.