Pith. sign in

REVIEW 3 major objections 5 minor 101 references

BPG: Balancing Plasticity and Generalization for Domain Incremental Learning

T0 review · 3 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

Pith's one-line read Domain incremental learning can reach near-zero forgetting by sizing each domain's adapter to its feature separability and softly mixing domain experts at test time, reaching 72.19% accuracy with 0.22% forgetting on DomainNet.

desk verdict Solid empirical paper with a useful plug-in; the capacity-allocation story is a heuristic dressed up as theory, but the experiments largely support the central claim. read the letter →

arxiv 2608.10804 v1 pith:YS5JUILE submitted 2026-08-11 cs.CV cs.AIcs.LG

classification cs.CVcs.AIcs.LG
keywords domainincrementallearningcatastrophicforgettingparameter-efficientfine-tuningadaptersfeatureseparabilitysoftmixturecontinualcapacityallocation
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 show that two persistent weaknesses of parameter-isolation domain-incremental learning can be fixed together: uniform adapter capacity ignores domain difficulty, and hard domain-ID selection at test time is brittle. It claims that giving harder domains larger adapter hidden dimensions and easier domains smaller ones, sized by a simple separability score, improves plasticity without wasting parameters, and combining this with a confidence-weighted soft mixture of domain-specific models at inference nearly eliminates forgetting. If true, rehearsal-free continual learning on pre-trained transformers improves substantially: BPG reports 72.19% average accuracy with 0.22% average forgetting on DomainNet with ViT-B/16, close to the oracle-domain-ID upper bound, with similar gains on CDDB and CORe50. The soft mixture is also plug-and-play, lifting accuracy when added to existing prompt-based methods.

What carries the argument

The load-bearing object is the feature-separability score $s_t=R^{\mathrm{bcs}}_t/(R^{\mathrm{wcs}}_t+\varepsilon)$, the ratio of between-class scatter to within-class scatter of frozen-backbone features for domain $t$. This feeds the capacity rule $r_t=(s_0/s_t)r_0$, which allocates a larger hidden dimension to domains whose classes are more entangled, with $s_0$ and $r_0$ fixed by a reference domain (ImageNet for separability; $r_0=64$ on DomainNet and CORe50, $r_0=1$ on CDDB). The supporting theoretical motivation is a qualitative risk decomposition $R_t(r_t)=A_t e^{-\kappa_t r_t}+B\sqrt{r_t/N_t}$, whose stationarity condition equates a marginal plasticity gain with a generalization and budget cost, yielding $r^*_t\propto 1/\kappa_t$ and, via a monotone map $\kappa_t\propto s_t$, the inverse rule. The second mechanism is BPG-Inference: $k=5$ k-means prototypes per domain, nearest-prototype distances $a_t=\min_j\|z-\nu_{t,j}\|_2$, softmax confidences $w_t$, a parameter-free pruning of domains below the uniform-confidence threshold, renormalization, and logit fusion $y_{\mathrm{final}}=\sum_t w'_t y_t$.

What would settle it

On a new multi-domain benchmark, compute each domain's separability score $s_t$ and then grid-search its optimal adapter dimension in isolation; if any high-separability domain peaks only at a large adapter dimension, or any low-separability domain peaks at a small one, the monotone inverse rule $r_t=(s_0/s_t)r_0$ fails.

Watch

Extended reading notes

Core claim

The paper's central claim is that the plasticity–generalization tension in parameter-isolation DIL is removable in a rehearsal-free setting. BPG-Adapter sets each domain's adapter hidden dimension by the inverse-proportionality rule $r_t=(s_0/s_t)r_0$, where $s_t=R^{\mathrm{bcs}}_t/(R^{\mathrm{wcs}}_t+\varepsilon)$ is a feature-separability ratio of between-class scatter to within-class scatter computed on frozen backbone features, and $s_0$ comes from ImageNet as the reference domain. BPG-Inference replaces the single-domain commitment with a soft mixture: per-domain k-means prototypes define a nearest-prototype affinity, a softmax turns distances into confidences, candidate domains below the uniform-confidence threshold are pruned, and the surviving per-domain classifiers are fused. On DomainNet with ViT-B/16 the framework reports 72.19% average accuracy and 0.22% average forgetting, within two points of the oracle-domain-ID upper bound; on CLIP it reports 75.72% with 0.59% forgetting, with analogous gains on CDDB and CORe50, and the soft mixture improves existing prompt methods when plugged into their inference.

Load-bearing premise

The paper assumes that a domain's feature-separability score reliably orders how much adapter capacity it needs, via an increasing map from separability to error-decay rate, and it hand-sets the reference capacity $r_0$ per dataset; if that ordering or the chosen reference is wrong, capacity is misallocated.

Editorial extensions

If this is right

  • On DomainNet with ViT-B/16, BPG reaches 72.19% average accuracy and 0.22% average forgetting, surpassing the next-best rehearsal-free method by 4.24% in accuracy and closing the gap to the oracle-domain-ID upper bound to 2.02 points.
  • On the CLIP backbone, BPG reaches 75.72% with 0.59% forgetting on DomainNet, 93.91% with 0.12% forgetting on CDDB, and 92.46% accuracy on CORe50, ahead of the corresponding prompt-based baselines.
  • BPG-Inference is a plug-and-play module: added on top of S-Prompts, PINA, C-Prompt, and KA-Prompt, it improves average accuracy on all three benchmarks, with the largest gain being +5.42 points on DomainNet.
  • BPG-Adapter transfers to LoRA as well as adapters, showing the capacity rule is not tied to a specific parameter-efficient fine-tuning module.
  • With $r_0=64$, BPG's total parameter budget matches a uniform adapter at $r=256$ yet improves both accuracy and forgetting, indicating that the gain comes from reallocating capacity rather than adding parameters.

Reading between the lines

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

  • If the monotone separability-to-capacity relationship holds generally, the hand-set reference capacity $r_0$ could itself be determined from the distribution of separability scores across the encountered domains, removing the last per-dataset hyperparameter; the paper does not explore this.
  • The large improvement from replacing hard selection with soft mixture suggests that much of what is reported as 'forgetting' in parameter-isolation methods may actually be domain-ID misclassification at test time rather than representation drift; reporting accuracy conditioned on correct domain assignment would disentangle the two sources.
  • Because BPG-Inference needs only frozen features and prototypes, it can be applied retrospectively to any already-trained parameter-isolation model; a natural stress test is to apply it to methods whose adapters were sized uniformly and measure how much of the gain comes from the soft mixture alone.
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

3 major / 5 minor

Summary. The paper proposes BPG, a parameter-isolation framework for domain incremental learning. BPG-Adapter sets each domain's adapter hidden dimension according to an inverse-proportionality rule based on a frozen-backbone feature separability score, while BPG-Inference replaces hard domain selection with a confidence-weighted soft mixture over domain-specific experts using k-means prototypes. Experiments on DomainNet, CDDB, and CORe50 with ViT-B/16 and CLIP backbones report state-of-the-art average accuracy and very low forgetting on several benchmarks, including A_T=72.19 and F_T=0.22 on DomainNet with ViT. The paper also shows that BPG-Inference can be applied as a plug-and-play module to existing prompt-based methods, and provides ablations, parameter-matched comparisons, and efficiency analyses.

Significance. If the results hold, BPG offers a simple and effective recipe for parameter-isolation DIL, and the soft-mixture inference module is a useful and broadly applicable contribution. The paper's experimental core is solid: ablations isolate the two components, the per-domain matrix in Table X directly illustrates the forgetting dynamics, and Section IV-G demonstrates gains over uniform scaling at matched parameter budgets. The paper also ships unusually detailed efficiency analyses (training time, latency, parameter counts) and makes the adaptive-capacity idea concrete via a reproducible separability score. The main weaknesses are the theoretical justification of the capacity rule, which relies on an unverified monotonicity assumption, and an apparent inconsistency between the printed per-domain matrix and the reported forgetting value.

major comments (3)
  1. [Section III-E and Section IV-F] The derivation of the capacity rule from Eq. (23) to Eq. (24) is not justified. The conclusion r* ∝ 1/κ_t ignores that A_t, N_t, and the logarithmic term in Eq. (23) all vary across domains, so the stated assumptions do not imply monotonicity of r* in κ_t. More importantly, the assumed monotone link κ_t ∝ s_t is load-bearing but only supported post-hoc by Fig. 4, which actually contains a counterexample: Sketch (s=0.156) peaks at r=256 while Quickdraw (s=0.151) peaks at r=1024, despite nearly identical separability scores. I recommend measuring κ_t directly (e.g., fitting per-domain risk curves) or demonstrating the monotone relationship on additional datasets before claiming that Eq. (14) approximates an optimal allocation.
  2. [Table X] The per-domain accuracy matrix reported in Table X does not reproduce the stated F_T values under the metric defined in Eq. (25). Recomputing from the printed entries for 'BPG-Adapter only' gives approximately F_T=1.59, not 1.28; for the full BPG matrix the recomputed value is about 0.24, close to but not equal to the reported 0.22. Please clarify the exact computation (e.g., any rounding, seed averaging, or a different forgetting definition) and provide matrix entries with enough precision so that the headline forgetting numbers are reproducible.
  3. [Section III-E.b and Section IV-G] The Lagrangian derivation in Eq. (21) assumes a fixed total budget R_tot, but Eq. (14) does not enforce any budget; the total capacity is an emergent function of s_0, r_0, and the sequence of s_t values. For low-separability domains, the rule can produce r_t exceeding the embedding dimension d, and the only safeguard in the paper is the per-dataset hand-setting of r_0 (64 for DomainNet/CORe50, 1 for CDDB). Please report the sensitivity of the final results to r_0 on CDDB and CORe50, and discuss how r_0 is selected without access to test performance, since this is a free parameter that directly controls the scale of all adapter widths.
minor comments (5)
  1. [Section IV-D.a and Fig. 3] The text states that BPG-Adapter is evaluated with r_0 = 4, 16, and 64, but Fig. 3 appears to show only one BPG-Adapter curve; please clarify whether the curve is a representative run, an average, or a different summary of the three settings.
  2. [Section III-D.b] The pruning step is described as using the uniform distribution over domains as a threshold; please state explicitly that this threshold is applied to the softmax-normalized weights w_t rather than to the raw distances a_t.
  3. [Table IX] The CORe50 row contains a malformed entry ('91.87–92.14–'); please fix the formatting so the table is readable.
  4. [Section IV-G.a] The adapter parameter count is given as 2L d r_t + L d + L r_t, but the adapter structure in Section III-B does not define which parameters correspond to the L d and L r_t terms; please specify whether these are biases or other parameters.
  5. [Section IV-F] The t-SNE visualizations in Fig. 4 are described in the text, but the figure is not referenced before Section IV-F; please add the reference where the per-domain accuracy numbers are first discussed.

Circularity Check

1 steps flagged · score 5.0 of 10

The capacity-allocation rule is justified by assuming κ_t∝s_t, and the same Fig. 4 trend serves as both motivation and corroboration; BPG-Inference is independently tested.

  1. self definitional [Section III-E.b (Eqs. 23–24) vs. Section III-C.b Eq. (14)]
    "This monotone relationship is corroborated by the empirical behavior reported in Section IV-F: the least separable domains (e.g., Infograph, s2=0.124, and Quickdraw, s4=0.151) keep benefiting from larger r up to r=1024, whereas the most separable one (Real, s5=0.530) already peaks at a small r and degrades under excess capacity; ... Adopting the simplest such map, κt ∝ st, Eq. (23) then reduces to the clean trend r⋆t ∝ 1/st, which is consistent with our capacity allocation rule in Eq. (14)."

    Eq. (14) was not derived from the toy risk model; it was imposed earlier through the multiplicative constraint λ=s_t r_t. The Section III-E derivation reaches r*∝1/s_t only after assuming an unmeasured monotone map κ_t=φ(s_t) and then 'adopting the simplest such map, κt∝st.' The only evidence offered for that map is the Section IV-F accuracy-vs-capacity behavior, which is the same phenomenon (easy domains peak at small r, hard domains keep improving up to large r) that motivated the inverse-proportionality rule in the first place (Fig. 1c). Thus Eq. (24) is effectively a restatement of the already-imposed Eq. (14), and the empirical 'corroboration' is the input trend, not an independent test. The assumed monotone link is also imperfect in the paper's own Fig.

full rationale

The only substantial circular element is the theoretical justification of BPG-Adapter. The capacity rule r_t=(s0/st)r0 is itself introduced as an imposed multiplicative constraint, and the later 'derivation' obtains the same inverse trend only by adopting κ_t∝s_t, whose support is the same Fig. 4 / Section IV-F behavior that motivated the rule. So the derivation is partly post-hoc rationalization rather than an independent derivation. The remainder of the paper is self-contained: BPG-Inference is validated against hard selection and against external prompt/parameter-isolation baselines, the ablations compare with uniform adapters, and the state-of-the-art accuracy/forgetting claims are empirical and could have failed. There is no load-bearing self-citation chain or imported uniqueness theorem. Because the central capacity mechanism's theoretical story reduces to its own input trend, but the experimental evaluation itself is not circular, a moderate score of 5 is appropriate.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The central mechanism rests on two externally supplied ingredients: a frozen pre-trained feature space (from ImageNet-pretrained ViT or CLIP) and the assumption that separability in that space is inversely related to the adapter capacity needed. The paper also introduces a hand-set reference dimension r0 per dataset. The soft-mixture inference adds a mild assumption that nearest-prototype distances in frozen features indicate which domain experts are reliable.

free parameters (2)
  • r0 (reference adapter hidden dimension) = 64 for DomainNet and CORe50, 1 for CDDB
    Hand-set per dataset in Section IV-A.d. It sets the overall capacity budget via r_t = (s0/st) r0. Changing r0 changes all per-domain dimensions and affects accuracy and forgetting, as shown in Fig. 3. No principled rule justifies 64 versus 1.
  • k (number of k-means prototypes per domain) = 5
    Used in BPG-Inference Eq (15) to build domain prototypes. No sensitivity analysis is reported; the choice affects domain affinity weights.
assumptions (4)
  • domain assumption Pre-trained ViT/CLIP features provide a valid estimate of domain difficulty via the separability score.
    Eq (11)-(13) compute s_t from frozen features; the entire BPG-Adapter rule depends on s_t monotonicity.
  • ad hoc to paper An increasing map kappa_t = phi(s_t) exists, and kappa_t proportional to s_t is the simplest valid choice.
    Section III-E.b states 'Adopting the simplest such map, kappa_t proportional to s_t, Eq. (23) then reduces to ...' This assumption is needed to obtain Eq (24); the paper supports it with the same empirical trend, Section IV-F, that the rule reproduces.
  • domain assumption The toy risk decomposition forms in Eq (18)-(19) capture the qualitative behavior of classification risk.
    Section III-E.a calls the expressions 'illustrative surrogates rather than exact laws'; the derivation of Eq (23) depends on their monotone structure.
  • domain assumption Nearest-prototype distance in the frozen feature space predicts which domain classifier is reliable for a test sample.
    Eq (15)-(16) define BPG-Inference weights; if distances do not track classifier competence, the soft mixture weights are wrong.

how reviews work

0 comments
Cite this review

Pith. "Pith review of BPG: Balancing Plasticity and Generalization for Domain Incremental Learning." pith.science (2026). https://pith.science/paper/YS5JUILE

@misc{pith2026260810804,
  author       = {Pith},
  title        = {Pith review of: BPG: Balancing Plasticity and Generalization for Domain Incremental Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YS5JUILE}},
  note         = {Machine review of arXiv:2608.10804}
}
read the original abstract

Deep neural networks excel in various tasks but struggle to generalize across evolving data distributions, leading to significant performance degradation under domain shifts. Domain incremental learning (DIL) addresses this challenge by enabling models to continuously adapt while retaining prior knowledge. Among existing DIL approaches, the parameter-isolation paradigm achieves state-of-the-art performance. However, these methods often adopt a one-size-fits-all approach to adapt to new domains, resulting in either insufficient learning capacity or redundant parameters. In this work, we propose BPG, a unified framework that addresses both challenges through two complementary components: BPG-Adapter, which dynamically determines each domain's adapter hidden dimension based on domain-specific feature separability, and BPG-Inference, a soft domain mixture strategy that integrates multiple domain-specific models at test time, mitigating domain ID misselection. Experimental results on DomainNet, CDDB, and CORe50 demonstrate that BPG consistently outperforms uniform adapter-based approaches and hard domain selection strategies, achieving state-of-the-art average accuracy while reducing forgetting to as low as 0.22% on DomainNet.

Figures

Figures reproduced from arXiv: 2608.10804 by the authors.

Figure 1
Figure 1. Motivation of BPG. (a) Domains from DomainNet exhibit varying levels of visual complexity. (b) t-SNE visualizations of pre-trained backbone (ViT￾B/16) features confirm that harder domains have lower feature separability scores (s1 > s2 > s3). (c) The optimal adapter dimension differs across domains: easier domains peak at smaller capacities while harder domains benefit from larger ones, motivating our adaptive alloc… view at source ↗
Figure 2
Figure 2. Overall architecture of the proposed BPG framework. An input image is tokenized by the patch embedding layer and concatenated with trainable [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Ablation study on the hidden dimension r and BPG-Adapter. The two curves compare models with and without BPG-Adapter under varying hidden dimensions, reporting both average accuracy AT (left) and average forgetting FT (right) on DomainNet. mainNet dataset, as illustrated in [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: t-SNE visualizations of domain-specific features on DomainNet under uniform adapter hidden dimensions. Each column is a domain and each row a [PITH_FULL_IMAGE:figures/full_fig_p010_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

101 extracted references · 49 canonical work pages

  1. [1]

    Query- aware cross-mixup and cross-reconstruction for few-shot fine-grained image classification,

    Z. Zhang, D. Chang, R. Zhu, X. Li, Z. Ma, and J.-H. Xue, “Query- aware cross-mixup and cross-reconstruction for few-shot fine-grained image classification,”IEEE Transactions on Circuits and Systems for Video Technology, vol. 35, no. 2, pp. 1276–1286, 2024. PREPRINT 12

  2. [2]

    Dynamic integration of task-specific adapters for class incremental learning,

    J. Li, S. Wang, B. Qian, Y . He, X. Wei, and Y . Gong, “Dynamic integration of task-specific adapters for class incremental learning,” arXiv preprint arXiv:2409.14983, 2024

  3. [3]

    Class-independent increment: An efficient approach for multi-label class-incremental learning,

    C. Ding, S. Dong, Z. Zhou, J. Han, Q. Wang, Y . He, and Y . Gong, “Class-independent increment: An efficient approach for multi-label class-incremental learning,”arXiv preprint arXiv:2503.00515, 2025

  4. [4]

    Learn by reasoning: Analogical weight generation for few-shot class- incremental learning,

    J. Han, C. Ding, Y . He, S. Dong, Q. Wang, X. Gao, and Y . Gong, “Learn by reasoning: Analogical weight generation for few-shot class- incremental learning,”IEEE Transactions on Circuits and Systems for Video Technology, 2025

  5. [5]

    Gfpl: Generative federated prototype learning for resource-constrained and data-imbalanced vision task,

    S. Lu, Y . He, J. Li, Q. Wang, and Y . Gong, “Gfpl: Generative federated prototype learning for resource-constrained and data-imbalanced vision task,”arXiv preprint arXiv:2602.21873, 2026

  6. [6]

    A robust moving object detection in multi-scenario big data for video surveillance,

    B.-H. Chen, L.-F. Shi, and X. Ke, “A robust moving object detection in multi-scenario big data for video surveillance,”IEEE Transactions on Circuits and Systems for Video Technology, vol. 29, no. 4, pp. 982–995, 2018

  7. [7]

    Learning endogenous attention for incremental object detection,

    X. Song, Y . He, J. Li, Q. Wang, and Y . Gong, “Learning endogenous attention for incremental object detection,” inProceedings of the Com- puter Vision and Pattern Recognition Conference, 2025, pp. 30 354– 30 364

  8. [8]

    Shared & domain self-adaptive experts with frequency-aware discrim- ination for continual test-time adaptation,

    J. Zhao, C. Ding, S. Dong, J. Li, Q. Wang, Y . He, and Y . Gong, “Shared & domain self-adaptive experts with frequency-aware discrim- ination for continual test-time adaptation,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 40, no. 34, 2026, pp. 28 697– 28 705

Show all 101 references
  1. [9]

    Decenter: Density-center guided perception enhancement for uav object detection,

    Z. Shi, Z. Wu, J. Wen, M. Li, X. Fan, Y . Wang, and L. Shen, “Decenter: Density-center guided perception enhancement for uav object detection,”IEEE Transactions on Circuits and Systems for Video Technology, 2026

  2. [10]

    Towards open-vocabulary video semantic segmentation,

    X. Li, Y . Liu, G. Sun, M. Wu, L. Zhang, and C. Zhu, “Towards open-vocabulary video semantic segmentation,”IEEE Transactions on Multimedia, 2025

  3. [11]

    Holistic prototype attention network for few-shot video object segmentation,

    Y . Tang, T. Chen, X. Jiang, Y . Yao, G.-S. Xie, and H.-T. Shen, “Holistic prototype attention network for few-shot video object segmentation,” IEEE Transactions on Circuits and Systems for Video Technology, vol. 34, no. 8, pp. 6699–6709, 2023

  4. [12]

    Catastrophic interference in connec- tionist networks: The sequential learning problem,

    M. McCloskey and N. J. Cohen, “Catastrophic interference in connec- tionist networks: The sequential learning problem,” inPsychology of learning and motivation. Elsevier, 1989, vol. 24, pp. 109–165

  5. [13]

    Lobg: less overfitting for better generalization in vision-language model,

    C. Ding, X. Gao, S. Dong, Y . He, Q. Wang, A. Kot, and Y . Gong, “Lobg: less overfitting for better generalization in vision-language model,”arXiv preprint arXiv:2410.10247, 2024

  6. [14]

    Goal: Geometrically optimal alignment for continual generalized cat- egory discovery,

    J. Han, C. Ding, S. Dong, Y . He, S. Wang, Q. Wang, and Y . Gong, “Goal: Geometrically optimal alignment for continual generalized cat- egory discovery,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 40, no. 6, 2026, pp. 4565–4573

  7. [15]

    Trajectory-diversity-driven robust vision-and-language navigation,

    J. Li, C. Wan, S. Dong, C. Ding, Q. Wang, Z. Ma, and Y . Gong, “Trajectory-diversity-driven robust vision-and-language navigation,” arXiv preprint arXiv:2603.15370, 2026

  8. [16]

    Consistent supervised-unsupervised alignment for general- ized category discovery,

    J. Han, S. Wang, Y . He, C. Ding, Q. Wang, X. Gao, S. Dong, and Y . Gong, “Consistent supervised-unsupervised alignment for general- ized category discovery,”arXiv preprint arXiv:2507.04725, 2025

  9. [17]

    Be- yond clip generalization: Against forward&backward forgetting adapter for continual learning of vision-language models,

    S. Dong, C. Ding, J. Li, J. Han, Q. Wang, Y . He, and Y . Gong, “Be- yond clip generalization: Against forward&backward forgetting adapter for continual learning of vision-language models,”arXiv preprint arXiv:2505.07690, 2025

  10. [18]

    Preventing catastrophic forgetting in continuous online learning for autonomous driving,

    R. Yang, T. Yang, Z. Yan, T. Krajnik, and Y . Ruichek, “Preventing catastrophic forgetting in continuous online learning for autonomous driving,” in2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2024, pp. 5505–5512

  11. [19]

    Dynamic memory to alleviate catastrophic forgetting in continual learning with medical imaging,

    M. Perkonigg, J. Hofmanninger, C. J. Herold, J. A. Brink, O. Pianykh, H. Prosch, and G. Langs, “Dynamic memory to alleviate catastrophic forgetting in continual learning with medical imaging,”Nature com- munications, vol. 12, no. 1, p. 5678, 2021

  12. [20]

    Continual learning for anomaly detection in surveillance videos,

    K. Doshi and Y . Yilmaz, “Continual learning for anomaly detection in surveillance videos,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition workshops, 2020, pp. 254– 255

  13. [21]

    Vdc- agent: When video detailed captioners evolve themselves via agentic self-reflection,

    Q. Wang, X. Gao, S. Dong, J. Han, J. Li, Y . He, and Y . Gong, “Vdc- agent: When video detailed captioners evolve themselves via agentic self-reflection,”arXiv preprint arXiv:2511.19436, 2025

  14. [22]

    S-prompts learning with pre-trained transformers: An occam’s razor for domain incremental learning,

    Y . Wang, Z. Huang, and X. Hong, “S-prompts learning with pre-trained transformers: An occam’s razor for domain incremental learning,” Advances in Neural Information Processing Systems, vol. 35, pp. 5682– 5695, 2022

  15. [23]

    Incremental learning of multi- domain image-to-image translations,

    D. S. Tan, Y .-X. Lin, and K.-L. Hua, “Incremental learning of multi- domain image-to-image translations,”IEEE Transactions on Circuits and Systems for Video Technology, vol. 31, no. 4, pp. 1526–1539, 2020

  16. [24]

    Beyond prompt learning: Continual adapter for efficient rehearsal-free continual learn- ing,

    X. Gao, S. Dong, Y . He, Q. Wang, and Y . Gong, “Beyond prompt learning: Continual adapter for efficient rehearsal-free continual learn- ing,” inEuropean Conference on Computer Vision. Springer, 2024, pp. 89–106

  17. [25]

    Video domain incremental learning for human action recognition in home environments,

    Y . Hu, J. Hou, X. Liu, X. Sun, and W. Guo, “Video domain incremental learning for human action recognition in home environments,” in International Conference on Image and Graphics. Springer, 2025, pp. 316–327

  18. [26]

    Class incremental learning for light-weighted networks,

    Z. Tao, L. Yu, H. Yao, S. Huang, and C. Xu, “Class incremental learning for light-weighted networks,”IEEE Transactions on Circuits and Systems for Video Technology, vol. 34, no. 12, pp. 12 210–12 220, 2024

  19. [27]

    Domain incremental object detection based on feature space topology preserving strategy,

    L. Ding, X. Song, Y . He, C. Wang, S. Dong, X. Wei, and Y . Gong, “Domain incremental object detection based on feature space topology preserving strategy,”IEEE Transactions on Circuits and Systems for Video Technology, vol. 34, no. 1, pp. 424–437, 2023

  20. [28]

    Inflora: Interference-free low-rank adaptation for continual learning,

    Y .-S. Liang and W.-J. Li, “Inflora: Interference-free low-rank adaptation for continual learning,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 23 638–23 647

  21. [29]

    Dc-lora: Domain correlation low-rank adaptation for domain incremental learning,

    L. Li, S. Wang, C. Li, Y . Yuan, and G. Wang, “Dc-lora: Domain correlation low-rank adaptation for domain incremental learning,” High-Confidence Computing, p. 100270, 2025

  22. [30]

    Selective experience replay for lifelong learn- ing,

    D. Isele and A. Cosgun, “Selective experience replay for lifelong learn- ing,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 32, no. 1, 2018

  23. [31]

    Expe- rience replay for continual learning,

    D. Rolnick, A. Ahuja, J. Schwarz, T. Lillicrap, and G. Wayne, “Expe- rience replay for continual learning,”Advances in Neural Information Processing Systems, vol. 32, 2019

  24. [32]

    Memory-efficient class- incremental learning for image classification,

    H. Zhao, H. Wang, Y . Fu, F. Wu, and X. Li, “Memory-efficient class- incremental learning for image classification,”IEEE Transactions on Neural Networks and Learning Systems, vol. 33, no. 10, pp. 5966– 5977, 2021

  25. [33]

    Overcoming catastrophic forgetting in neural networks,

    J. Kirkpatrick, R. Pascanu, N. Rabinowitz, J. Veness, G. Desjardins, A. A. Rusu, K. Milan, J. Quan, T. Ramalho, A. Grabska-Barwinska et al., “Overcoming catastrophic forgetting in neural networks,”Pro- ceedings of the national academy of sciences, vol. 114, no. 13, pp. 3521–3526, 2017

  26. [34]

    Continual learning through synaptic intelligence,

    F. Zenke, B. Poole, and S. Ganguli, “Continual learning through synaptic intelligence,” inInternational conference on machine learning. PMLR, 2017, pp. 3987–3995

  27. [35]

    Memory aware synapses: Learning what (not) to forget,

    R. Aljundi, F. Babiloni, M. Elhoseiny, M. Rohrbach, and T. Tuytelaars, “Memory aware synapses: Learning what (not) to forget,” inProceed- ings of the European conference on computer vision (ECCV), 2018, pp. 139–154

  28. [36]

    Subspace regularizers for few-shot class incremental learning,

    A. F. Aky ¨urek, E. Aky ¨urek, D. T. Wijaya, and J. Andreas, “Subspace regularizers for few-shot class incremental learning,”arXiv preprint arXiv:2110.07059, 2021

  29. [37]

    Few-shot class-incremental learning via entropy-regularized data-free replay,

    H. Liu, L. Gu, Z. Chi, Y . Wang, Y . Yu, J. Chen, and J. Tang, “Few-shot class-incremental learning via entropy-regularized data-free replay,” in European Conference on Computer Vision. Springer, 2022, pp. 146– 162

  30. [38]

    Multi-granularity knowledge distillation and prototype consistency regularization for class-incremental learning,

    Y . Shi, D. Shi, Z. Qiao, Z. Wang, Y . Zhang, S. Yang, and C. Qiu, “Multi-granularity knowledge distillation and prototype consistency regularization for class-incremental learning,”Neural Networks, vol. 164, pp. 617–630, 2023

  31. [39]

    Mop-clip: A mixture of prompt-tuned clip models for domain incre- mental learning,

    J. Nicolas, F. Chiaroni, I. Ziko, O. Ahmad, C. Desrosiers, and J. Dolz, “Mop-clip: A mixture of prompt-tuned clip models for domain incre- mental learning,” inProceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, 2024, pp. 1762–1772

  32. [40]

    Non- exemplar domain incremental learning via cross-domain concept in- tegration,

    Q. Wang, Y . He, S. Dong, X. Gao, S. Wang, and Y . Gong, “Non- exemplar domain incremental learning via cross-domain concept in- tegration,” inEuropean Conference on Computer Vision. Springer, 2024, pp. 144–162

  33. [41]

    Compositional prompting for anti- forgetting in domain incremental learning,

    Z. Liu, Y . Peng, and J. Zhou, “Compositional prompting for anti- forgetting in domain incremental learning,”International Journal of Computer Vision, pp. 1–18, 2024

  34. [42]

    Importance- aware shared parameter subspace learning for domain incremental learning,

    S. Wang, C. Li, J. Tang, X. Gong, Y . Yuan, and G. Wang, “Importance- aware shared parameter subspace learning for domain incremental learning,” inProceedings of the 32nd ACM International Conference on Multimedia, 2024, pp. 8874–8883

  35. [43]

    Imbalanced continual learning with partitioning reservoir sampling,

    C. D. Kim, J. Jeong, and G. Kim, “Imbalanced continual learning with partitioning reservoir sampling,” inComputer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part XIII 16. Springer, 2020, pp. 411–428. PREPRINT 13

  36. [44]

    Space rotation with basis transformation for training-free test-time adaptation,

    C. Ding, X. Gao, S. Dong, Y . He, Q. Wang, X. Song, A. Kot, and Y . Gong, “Space rotation with basis transformation for training-free test-time adaptation,”arXiv preprint arXiv:2502.19946, 2025

  37. [45]

    Vari- ational prototype replays for continual learning,

    M. Zhang, T. Wang, J. H. Lim, G. Kreiman, and J. Feng, “Vari- ational prototype replays for continual learning,”arXiv preprint arXiv:1905.09447, 2019

  38. [46]

    Continual learning with deep generative replay,

    H. Shin, J. K. Lee, J. Kim, and J. Kim, “Continual learning with deep generative replay,”Advances in neural information processing systems, vol. 30, 2017

  39. [47]

    Class-incremental learning with generative classifiers,

    G. M. Van De Ven, Z. Li, and A. S. Tolias, “Class-incremental learning with generative classifiers,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021, pp. 3611–3620

  40. [48]

    Ranpac: Random projections and pre-trained mod- els for continual learning,

    M. D. McDonnell, D. Gong, A. Parvaneh, E. Abbasnejad, and A. Van den Hengel, “Ranpac: Random projections and pre-trained mod- els for continual learning,”Advances in Neural Information Processing Systems, vol. 36, pp. 12 022–12 053, 2023

  41. [49]

    Revisiting class-incremental learning with pre-trained models: Generalizability and adaptivity are all you need,

    D.-W. Zhou, Z.-W. Cai, H.-J. Ye, D.-C. Zhan, and Z. Liu, “Revisiting class-incremental learning with pre-trained models: Generalizability and adaptivity are all you need,”International Journal of Computer Vision, vol. 133, no. 3, pp. 1012–1032, 2025

  42. [50]

    Slca: Slow learner with classifier alignment for continual learning on a pre-trained model,

    G. Zhang, L. Wang, G. Kang, L. Chen, and Y . Wei, “Slca: Slow learner with classifier alignment for continual learning on a pre-trained model,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 19 148–19 158

  43. [52]

    Is parameter isolation better for prompt-based continual learning?

    J. Li, C. Ding, S. Dong, Q. Wang, J. Zhao, Y . He, and Y . Gong, “Is parameter isolation better for prompt-based continual learning?”arXiv preprint arXiv:2601.20894, 2026

  44. [53]

    Privacy-preserving continual learning methods for medical image classification: a comparative analysis,

    T. Verma, L. Jin, J. Zhou, J. Huang, M. Tan, B. C. M. Choong, T. F. Tan, F. Gao, X. Xu, D. S. Tinget al., “Privacy-preserving continual learning methods for medical image classification: a comparative analysis,” Frontiers in Medicine, vol. 10, p. 1227515, 2023

  45. [54]

    Preventing zero-shot transfer degradation in continual learning of vision-language models,

    Z. Zheng, M. Ma, K. Wang, Z. Qin, X. Yue, and Y . You, “Preventing zero-shot transfer degradation in continual learning of vision-language models,” inProceedings of the IEEE/CVF international conference on computer vision, 2023, pp. 19 125–19 136

  46. [55]

    Continual learning and catastrophic forgetting,

    G. M. Van de Ven, N. Soures, and D. Kudithipudi, “Continual learning and catastrophic forgetting,”arXiv preprint arXiv:2403.05175, 2024

  47. [56]

    Componential prompt-knowledge alignment for domain incremental learning,

    K. Xu, X. Zou, G. Hua, and J. Zhou, “Componential prompt-knowledge alignment for domain incremental learning,” inInternational Confer- ence on Machine Learning. PMLR, 2025, pp. 70 032–70 046

  48. [57]

    A unified continual learning framework with general parameter-efficient tuning,

    Q. Gao, C. Zhao, Y . Sun, T. Xi, G. Zhang, B. Ghanem, and J. Zhang, “A unified continual learning framework with general parameter-efficient tuning,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 11 483–11 493

  49. [58]

    Expandable sub- space ensemble for pre-trained model-based class-incremental learn- ing,

    D.-W. Zhou, H.-L. Sun, H.-J. Ye, and D.-C. Zhan, “Expandable sub- space ensemble for pre-trained model-based class-incremental learn- ing,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 23 554–23 564

  50. [59]

    Learning multiple visual domains with residual adapters,

    S.-A. Rebuffi, H. Bilen, and A. Vedaldi, “Learning multiple visual domains with residual adapters,”Advances in neural information processing systems, vol. 30, 2017

  51. [60]

    Lora: Low-rank adaptation of large language models

    E. J. Hu, Y . Shen, P. Wallis, Z. Allen-Zhu, Y . Li, S. Wang, L. Wang, W. Chenet al., “Lora: Low-rank adaptation of large language models.” ICLR, vol. 1, no. 2, p. 3, 2022

  52. [61]

    Parameter-efficient transfer learning for nlp,

    N. Houlsby, A. Giurgiu, S. Jastrzebski, B. Morrone, Q. De Laroussilhe, A. Gesmundo, M. Attariyan, and S. Gelly, “Parameter-efficient transfer learning for nlp,” inInternational conference on machine learning. PMLR, 2019, pp. 2790–2799

  53. [62]

    Adalora: Adaptive budget allocation for parameter-efficient fine-tuning,

    Q. Zhang, M. Chen, A. Bukharin, N. Karampatziakis, P. He, Y . Cheng, W. Chen, and T. Zhao, “Adalora: Adaptive budget allocation for parameter-efficient fine-tuning,”arXiv preprint arXiv:2303.10512, 2023

  54. [63]

    Dylora: Parameter efficient tuning of pre-trained models using dynamic search- free low-rank adaptation,

    M. Valipour, M. Rezagholizadeh, I. Kobyzev, and A. Ghodsi, “Dylora: Parameter efficient tuning of pre-trained models using dynamic search- free low-rank adaptation,”arXiv preprint arXiv:2210.07558, 2022

  55. [64]

    Gelora: Geo- metric adaptive ranks for efficient lora fine-tuning,

    A. Ed-dib, Z. Datbayev, and A. M. Aboussalah, “Gelora: Geo- metric adaptive ranks for efficient lora fine-tuning,”arXiv preprint arXiv:2412.09250, vol. 3, 2024

  56. [65]

    Ard-lora: Dynamic rank allocation for parameter-efficient fine-tuning of foundation models with heteroge- neous adaptation needs,

    H. U. K. Shinwari and M. Usama, “Ard-lora: Dynamic rank allocation for parameter-efficient fine-tuning of foundation models with heteroge- neous adaptation needs,”IEEE Transactions on Artificial Intelligence, 2025

  57. [66]

    Adaptive adapters: An efficient way to incorporate bert into neural machine translation,

    J. Guo, Z. Zhang, L. Xu, B. Chen, and E. Chen, “Adaptive adapters: An efficient way to incorporate bert into neural machine translation,” IEEE/ACM Transactions on Audio, Speech, and Language Processing, vol. 29, pp. 1740–1751, 2021

  58. [67]

    Dynamic adapter meets prompt tuning: Parameter-efficient transfer learning for point cloud analysis,

    X. Zhou, D. Liang, W. Xu, X. Zhu, Y . Xu, Z. Zou, and X. Bai, “Dynamic adapter meets prompt tuning: Parameter-efficient transfer learning for point cloud analysis,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 14 707–14 717

  59. [68]

    Adaptive adapter routing for long-tailed class-incremental learning,

    Z.-H. Qi, D.-W. Zhou, Y . Yao, H.-J. Ye, and D.-C. Zhan, “Adaptive adapter routing for long-tailed class-incremental learning,”Machine Learning, vol. 114, no. 3, pp. 1–20, 2025

  60. [69]

    Isolation and impartial aggregation: A paradigm of incremental learning without interference,

    Y . Wang, Z. Ma, Z. Huang, Y . Wang, Z. Su, and X. Hong, “Isolation and impartial aggregation: A paradigm of incremental learning without interference,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 37, no. 8, 2023, pp. 10 209–10 217

  61. [70]

    Lifelong language pretraining with distribution-specialized experts,

    W. Chen, Y . Zhou, N. Du, Y . Huang, J. Laudon, Z. Chen, and C. Cui, “Lifelong language pretraining with distribution-specialized experts,” inInternational Conference on Machine Learning. PMLR, 2023, pp. 5383–5395

  62. [71]

    Boosting continual learning of vision-language models via mixture-of-experts adapters,

    J. Yu, Y . Zhuge, L. Zhang, P. Hu, D. Wang, H. Lu, and Y . He, “Boosting continual learning of vision-language models via mixture-of-experts adapters,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 23 219–23 230

  63. [72]

    Moe-adapters++: Towards more efficient continual learning of vision-language models via dynamic mixture-of-experts adapters,

    J. Yu, Z. Huang, Y . Zhuge, L. Zhang, P. Hu, D. Wang, H. Lu, and Y . He, “Moe-adapters++: Towards more efficient continual learning of vision-language models via dynamic mixture-of-experts adapters,” IEEE Transactions on Pattern Analysis and Machine Intelligence, 2025

  64. [73]

    Imagenet: A large-scale hierarchical image database,

    J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li, and L. Fei-Fei, “Imagenet: A large-scale hierarchical image database,” in2009 IEEE conference on computer vision and pattern recognition. Ieee, 2009, pp. 248–255

  65. [74]

    Shalev-Shwartz and S

    S. Shalev-Shwartz and S. Ben-David,Understanding machine learning: From theory to algorithms. Cambridge university press, 2014

  66. [75]

    Mohri, A

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

  67. [76]

    Moment matching for multi-source domain adaptation,

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

  68. [77]

    A continual deepfake detection benchmark: Dataset, methods, and essentials,

    C. Li, Z. Huang, D. P. Paudel, Y . Wang, M. Shahbazi, X. Hong, and L. Van Gool, “A continual deepfake detection benchmark: Dataset, methods, and essentials,” inProceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, 2023, pp. 1339–1349

  69. [78]

    Core50: a new dataset and benchmark for continuous object recognition,

    V . Lomonaco and D. Maltoni, “Core50: a new dataset and benchmark for continuous object recognition,” inConference on robot learning. PMLR, 2017, pp. 17–26

  70. [79]

    An image is worth 16x16 words: Transformers for image recognition at scale,

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

  71. [80]

    Learning transferable visual models from natural language supervision,

    A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal, G. Sastry, A. Askell, P. Mishkin, J. Clarket al., “Learning transferable visual models from natural language supervision,” inInternational conference on machine learning. PMLR, 2021, pp. 8748–8763

  72. [81]

    Dytox: Trans- formers for continual learning with dynamic token expansion,

    A. Douillard, A. Ram ´e, G. Couairon, and M. Cord, “Dytox: Trans- formers for continual learning with dynamic token expansion,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 9285–9295

  73. [82]

    Learning without forgetting,

    Z. Li and D. Hoiem, “Learning without forgetting,”IEEE transactions on pattern analysis and machine intelligence, vol. 40, no. 12, pp. 2935– 2947, 2017

  74. [83]

    Learning to prompt for continual learning,

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

  75. [84]

    Dualprompt: Complementary prompting for rehearsal-free continual learning,

    Z. Wang, Z. Zhang, S. Ebrahimi, R. Sun, H. Zhang, C.-Y . Lee, X. Ren, G. Su, V . Perot, J. Dyet al., “Dualprompt: Complementary prompting for rehearsal-free continual learning,” inEuropean Conference on Computer Vision. Springer, 2022, pp. 631–648

  76. [85]

    Coda-prompt: Continual decomposed attention-based prompting for rehearsal-free continual learning,

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

  77. [86]

    Dual consolidation for pre-trained model-based domain-incremental learn- PREPRINT 14 ing,

    D.-W. Zhou, Z.-W. Cai, H.-J. Ye, L. Zhang, and D.-C. Zhan, “Dual consolidation for pre-trained model-based domain-incremental learn- PREPRINT 14 ing,” inProceedings of the Computer Vision and Pattern Recognition Conference, 2025, pp. 20 547–20 557

  78. [87]

    Dualcp: Rehearsal-free domain-incremental learning via dual-level concept prototype,

    Q. Wang, Y . He, S. Dong, X. Song, J. Han, H. Luo, and Y . Gong, “Dualcp: Rehearsal-free domain-incremental learning via dual-level concept prototype,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 39, no. 20, 2025, pp. 21 198–21 206

  79. [88]

    Boosting domain incremental learning: Selecting the optimal parame- ters is all you need,

    Q. Wang, X. Song, Y . He, J. Han, C. Ding, X. Gao, and Y . Gong, “Boosting domain incremental learning: Selecting the optimal parame- ters is all you need,” inProceedings of the Computer Vision and Pattern Recognition Conference, 2025, pp. 4839–4849

  80. [89]

    Continual knowledge consolidation lora for domain incremental learning,

    N. Paeedeh, M. Pratama, W. Ding, J. Cao, W. Mayer, and R. Kowal- czyk, “Continual knowledge consolidation lora for domain incremental learning,”arXiv preprint arXiv:2510.16077, 2025

  81. [90]

    Addressing imbal- anced domain-incremental learning through dual-balance collaborative experts,

    L. Li, D.-W. Zhou, H.-J. Ye, and D.-C. Zhan, “Addressing imbal- anced domain-incremental learning through dual-balance collaborative experts,” inInternational Conference on Machine Learning. PMLR, 2025, pp. 36 974–36 992

  82. [91]

    Prompt customization for continual learning,

    Y . Dai, X. Hong, Y . Wang, Z. Ma, D. Jiang, and Y . Wang, “Prompt customization for continual learning,”IEEE Transactions on Artificial Intelligence, 2025

  83. [92]

    Versatile incremental learning: Towards class and domain-agnostic incremental learning,

    M.-Y . Park, J.-H. Lee, and G.-M. Park, “Versatile incremental learning: Towards class and domain-agnostic incremental learning,” inEuropean Conference on Computer Vision. Springer, 2024, pp. 271–288

  84. [93]

    Hierar- chical decomposition of prompt-based continual learning: Rethinking obscured sub-optimality,

    L. Wang, J. Xie, X. Zhang, M. Huang, H. Su, and J. Zhu, “Hierar- chical decomposition of prompt-based continual learning: Rethinking obscured sub-optimality,”Advances in Neural Information Processing Systems, vol. 36, pp. 69 054–69 076, 2023

  85. [94]

    Cp-prompt: Composition-based cross-modal prompting for domain- incremental continual learning,

    Y . Feng, Z. Tian, Y . Zhu, Z. Han, H. Luo, G. Zhang, and M. Song, “Cp-prompt: Composition-based cross-modal prompting for domain- incremental continual learning,” inProceedings of the 32nd ACM International Conference on Multimedia, 2024, pp. 2729–2738

  86. [95]

    Latent replay for real-time continual learning,

    L. Pellegrini, G. Graffieti, V . Lomonaco, and D. Maltoni, “Latent replay for real-time continual learning,” in2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2020, pp. 10 203–10 209

  87. [96]

    Incremental learning for the detection and classification of gan-generated images,

    F. Marra, C. Saltori, G. Boato, and L. Verdoliva, “Incremental learning for the detection and classification of gan-generated images,” in2019 IEEE international workshop on information forensics and security (WIFS). IEEE, 2019, pp. 1–6

  88. [97]

    Learning a unified classifier incrementally via rebalancing,

    S. Hou, X. Pan, C. C. Loy, Z. Wang, and D. Lin, “Learning a unified classifier incrementally via rebalancing,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2019, pp. 831–839

  89. [98]

    On tiny episodic memories in continual learning,

    A. Chaudhry, M. Rohrbach, M. Elhoseiny, T. Ajanthan, P. K. Dokania, P. H. Torr, and M. Ranzato, “On tiny episodic memories in continual learning,”arXiv preprint arXiv:1902.10486, 2019

  90. [99]

    Gdumb: A simple approach that questions our progress in continual learning,

    A. Prabhu, P. H. Torr, and P. K. Dokania, “Gdumb: A simple approach that questions our progress in continual learning,” inComputer Vision– ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part II 16. Springer, 2020, pp. 524–540

  91. [100]

    Large scale incremental learning,

    Y . Wu, Y . Chen, L. Wang, Y . Ye, Z. Liu, Y . Guo, and Y . Fu, “Large scale incremental learning,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2019, pp. 374–382

  92. [101]

    Dark experience for general continual learning: a strong, simple baseline,

    P. Buzzega, M. Boschini, A. Porrello, D. Abati, and S. Calderara, “Dark experience for general continual learning: a strong, simple baseline,”Advances in neural information processing systems, vol. 33, pp. 15 920–15 930, 2020

  93. [102]

    Co2l: Contrastive continual learning,

    H. Cha, J. Lee, and J. Shin, “Co2l: Contrastive continual learning,” in Proceedings of the IEEE/CVF International conference on computer vision, 2021, pp. 9516–9525

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.