Pith. sign in

REVIEW 4 major objections 6 minor 44 references

The paper identifies a 'regional shortcut' — base training's focus on only the most discriminative image regions — as the cause of novel-class misclassification in few-shot class-incremental learning, and proposes a two-set primitive learni

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-01 05:53 UTC pith:M6Q3BRWM

load-bearing objection The empirical method is real and the two-primitive-bank idea is worth a look, but the paper's causal theory of regional shortcuts is circular and the validation doesn't test the load-bearing assumption. the 4 major comments →

arxiv 2607.22072 v1 pith:M6Q3BRWM submitted 2026-07-24 cs.CV

Alleviating Regional Shortcuts for Few-Shot Class-Incremental Learning

Haichen Zhou , Yazhe Lyu , Yixiong Zou , Ruixuan Li , Yuhua Li This is my paper
classification cs.CV
keywords few-shot class-incremental learningregional shortcutvisual primitivescompositional learningbase-novel confusioncommon primitive setinterpretabilitycatastrophic forgetting
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper tries to establish why few-shot class-incremental learning (FSCIL) models so often misclassify novel-class samples as base classes, and to fix it. It argues that during base-class training, cross-entropy loss drives a model to focus on only the most discriminative local image regions — 'primitives' — and to suppress everything else, a bias it calls the regional shortcut. That bias carries over to novel classes, so the model keeps attending to base-class-discriminative regions and fails to shift attention to novel-relevant ones. The paper formalizes this with a theorem: under a stated assumption, any local minimizer of the base training loss drives the weights of all non-discriminative primitives to zero. It then proposes ARS-CDP, which learns two primitive sets — a common set shared across classes and a discriminative set for class-specific cues — and constrains the model to use both, reporting consistent accuracy gains and clearer attention maps on standard benchmarks.

Core claim

The paper's central claim is that the 'regional shortcut' — the tendency of base-class cross-entropy training to concentrate the model on a small set of maximally discriminative regions (primitives) — is the mechanism behind base-novel confusion in FSCIL. The formal result, Theorem 1, states that under Assumption 1 (Non-Degradation of Competing Classes), any local minimizer of the cosine cross-entropy loss drives the weights of all non-discriminative primitives to zero. Consequently, the learned primitive set lacks semantic diversity, and when transferred to novel classes it over-weights base-class-discriminative regions and under-weights novel-relevant regions. To break this, the paper prop

What carries the argument

Visual primitives are the local spatial vectors of the backbone's feature map, treated as reusable units for composing class representations; a primitive counts as activated if its attention score exceeds a threshold. The 'regional shortcut' is the mechanism formalized by Theorem 1: under gradient dynamics, weights of non-discriminative primitives go to zero, so the model learns an overly narrow set of discriminative regions. ARS-CDP's machinery is the two learned primitive sets — a common set R_u (class-agnostic, shared semantic cues) and a discriminative set R_i (class-specific features) — constructed in a first stage and fixed in a second stage, with a similarity-guided loss and a prototy

Load-bearing premise

The load-bearing premise is Assumption 1 — that moving representation weight from any non-discriminative region to a discriminative region never helps any competing class; the paper's own verification only measures logit margins, not this pairwise condition, and the theoretical model also assumes learnable per-primitive weights while the experiments use uniform pooling, so if either premise fails the collapse argument and the mechanism claim weaken.

What would settle it

Directly measure the pairwise condition in Assumption 1 on a trained base model: for each non-discriminative primitive i and discriminative primitive s, compute ⟨w_k − w_y, f_s − f_i⟩ for every competing class k; if any value is positive, the assumption fails and Theorem 1's conclusion is not guaranteed. Also, retrain the baseline with learnable per-primitive weights a_i instead of uniform pooling and check whether primitive activation ratios on novel classes remain low; if they become diverse, the regional shortcut is an artifact of pooling choice rather than the loss dynamics described.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Share X Bluesky LinkedIn Reddit HN

If this is right

  • If the regional shortcut is the cause of base-novel confusion, then freezing a base-trained feature extractor for novel-class learning inherits this bias; debiasing requires constraining the primitive set, not just more data or augmentation.
  • Explicitly separating common and discriminative primitives should reduce novel-to-base misclassification and improve novel-class accuracy across incremental sessions, as the paper reports.
  • The two-set decomposition provides interpretable attention: common primitives capture shared textures and backgrounds, discriminative primitives capture fine-grained class details.
  • The reported gains on both ResNet and ViT backbones suggest the shortcut is a general phenomenon of cosine cross-entropy training, not an artifact of one architecture.
  • Theorem 1 implies that any method that maximizes cosine cross-entropy without diversity constraints will tend to collapse attention, so adding a common-primitive constraint is a general design principle for transferable representations.

Where Pith is reading between the lines

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

  • The paper's Assumption 1 is nearly a restatement of the theorem's conclusion; a direct test would measure the pairwise inequality ⟨w_k − w_y, f_s − f_i⟩ ≤ 0 on trained models, since Table IV's logit-margin comparison does not verify it.
  • The theory assumes learnable per-primitive weights a_i, while the experimental baseline uses uniform average pooling; if the baseline were given learnable weights, the primitive activation ratios might already change, which would test whether the analyzed gradient mechanism is the operative one.
  • If regional shortcuts are real, they should appear in other transfer settings (zero-shot, domain adaptation) as reduced activation diversity; measuring primitive activation ratios there would be a cheap out-of-distribution check.
  • A simpler remedy than two primitive sets might be a diversity or orthogonality penalty on attention weights; the paper's ablations show both sets help, but do not compare against such a penalty.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper addresses the tendency of few-shot class-incremental learning (FSCIL) models to misclassify novel-class samples as base classes. It attributes this to a 'regional shortcut' formed during base-class training: the model over-focuses on base-class-discriminative primitives and fails to acquire reusable common primitives. The authors provide a compositional interpretation, formalize the shortcut with a weighted-sum primitive model and an assumption, and prove a theorem stating that local minima concentrate on a discriminative primitive set. They then propose ARS-CDP, which constructs a common primitive set and a discriminative primitive set, and uses losses L_sim, L_pro, and L_r to encourage use of the common set. Experiments on CIFAR-100, miniImageNet, and CUB-200 report consistent improvements over state-of-the-art methods, with ablations showing that each component contributes.

Significance. If the regional-shortcut mechanism and ARS-CDP are valid, the paper would provide a useful explanation of a known FSCIL failure mode and a practical remedy. The empirical package is a genuine strength: Tables II, V, VI, and VII show that the proposed losses and primitive sets improve over baselines, and Table I reports consistent gains on CIFAR-100 with two backbones. However, the theoretical identification is not established. Assumption 1 (Eq. 6) nearly restates Theorem 1, the purported verification in Table IV tests a different quantity, and the analyzed optimization dynamics do not match the fixed average-pooling baseline used in the experiments. The empirical gains may stand independently, but the causal claim in the abstract and Section III-D is currently unsupported by the theoretical apparatus.

major comments (4)
  1. [Sec. III-C.3, Eq. (6)] Assumption 1 is effectively the theorem. Eq. (6) states that for every competitor k, replacing a non-discriminative primitive i by a discriminative primitive s cannot increase the relative logit of k over y. Since S is defined by <w_y,f_s> >= <w_y,f_i>, this condition already ensures that moving weight from i to s never increases the loss; substituting it into Eq. (11) gives Proposition 1 and then Theorem 1 in one line. The proof is therefore a restatement of the assumption. Table IV does not verify Eq. (6): it reports the global logit margin z_y - max_k z_k, not the per-primitive pairwise inequalities. To support the causal claim, the authors should either derive Eq. (6) from a weaker condition on trained models, or directly measure the pairwise quantities in Eq. (6) for a trained baseline and report how often they hold.
  2. [Sec. III-C.3, Theorem 1] Theorem 1 as stated is false under the stated assumptions. If Eq. (6) holds with equality for all k for some pair (i,s), then the loss is invariant to transferring weight between i and s, so a local minimizer can have a_i^*>0. The conclusion a_j^*->0 requires strict inequality for at least one competing class for every pair. The proposition's 'moreover' clause is not carried into the theorem. Please add the strictness condition explicitly and either prove the stronger claim or weaken 'any local minimizer' to 'there exists a minimizer.'
  3. [Sec. III-C.3 vs. Sec. III-A/Eq. (3)] The theory models phi(x)=sum_i a_i f_i with learnable nonnegative weights a_i and differentiates with respect to a_i. The baseline in Eq. (3) uses fixed average pooling (a_i=1/N) inside a cosine classifier; no per-primitive weights are learned. Thus the gradient mechanism analyzed in Proposition 1 and Theorem 1 is not the optimization dynamics of the evaluated model. The sentence in Sec. III-C.3 claiming general applicability to visual backbones does not bridge this gap. Either analyze the actual average-pooling formulation or explicitly present the result as an analogy/hypothesis rather than a derivation for the experimental setting.
  4. [Abstract and Sec. III-D] The paper's central claim is causal: base-class CE training creates regional shortcuts, and these cause novel-base confusion. The empirical evidence in Figs. 3 and 4 shows correlation (models with low primitive activation also misclassify novel as base), and the ablations in Tables II and VII show that the proposed intervention improves accuracy. These results are consistent with the shortcut hypothesis but do not isolate the causal chain. In particular, Table II shows that L_r drives most of the novel-class gain (23.40 to 29.97), so the improvement may be attributable to the reconstruction-style loss rather than specifically to the common/discriminative split. Please add an intervention that directly manipulates the hypothesized shortcut (e.g., enforcing or suppressing primitive diversity) and measures the effect on novel-base confusion, or temper the causal wording.
minor comments (6)
  1. [Eqs. (4) and (8)] The loss in Eq. (4) uses tau*cos(phi,w), but Lemma 2 in Eq. (8) uses exp(<w_k,phi>) without temperature or normalization. Reconcile the notation.
  2. [Sec. III-C.1] The threshold for a primitive to be 'activated' is not specified. Please define it, since the activation-ratio analyses in Figs. 3 and 4 depend on it.
  3. [Table I] The R-FSCIL row appears corrupted ('ResNet-1282.9076.30...'); fix the formatting. Also clarify whether 'surpasses' refers to average accuracy, since the session-0 accuracy of Ours is lower than some baselines.
  4. [Fig. 6] MiniImageNet and CUB-200 are only shown as plots. Provide numeric per-session and average accuracies for reproducibility and easier comparison.
  5. [Sec. V.C] The values of lambda and beta used in the main experiments are not reported; the sensitivity plot alone is insufficient for reproduction.
  6. [Sec. III-C.3, Theorem 1] The notation a_j^* -> 0 suggests an asymptotic limit; state the result as an equality a_j^*=0 at the minimizer.

Circularity Check

1 steps flagged

Assumption 1 (Eq. 6) already asserts the 'regional shortcut'; Theorem 1 restates it via one gradient sign check, and the reported validation tests margins, not the pairwise condition.

specific steps
  1. other [Sec. III-C.3, Assumption 1 (Eq. 6), Proposition 1 (Eq. 11), Theorem 1 (Eq. 12)]
    "Assumption 1. Non-Degradation of Competing Classes. For any non-discriminative primitive i /∈ S and any discriminative primitive s ∈ S, transferring weight from i to s does not increase the activation or logit of any competing class k ̸= y. In other words, the discriminative primitives in S are universally superior for the ground-truth class y, and reallocating representation mass from outside S to S cannot improve the relative advantage of any alternative class. Formally, this condition can be expressed as ⟨wk − wy, fs − fi⟩ ≤ 0, ∀i /∈ S, s ∈ S, k ̸= y, (6)"

    The theorem's conclusion (non-discriminative weights a*_j → 0) is derived by taking Eq. (6) as the premise in Proposition 1: ∂L/∂a_s − ∂L/∂a_i = Σ_k p_k⟨w_k−w_y, f_s−f_i⟩ ≤ 0. But Eq. (6) already says discriminative primitives dominate non-discriminative ones for the true class against every competitor — i.e. it asserts the 'regional shortcut' the paper claims to prove. Theorem 1 is one sign check away from Assumption 1, so it restates the premise rather than deriving the mechanism. Table IV validates the assumption using z_y − max_k z_k, a global logit margin, not the pairwise inequality of Eq. (6); the load-bearing premise is not independently anchored.

full rationale

The empirical parts of the paper are not circular: ARS-CDP is evaluated on standard FSCIL benchmarks and its gains could in principle be real, with ablations against data augmentation and mask strategies. The circularity is localized to the theory supporting the causal claim. The paper defines S as the primitives maximally aligned with w_y, then assumes (Eq. 6) that every s∈S has a weakly better gradient direction than every i∉S for all competitors; Proposition 1 and Theorem 1 follow by plugging this assumption into the CE gradient. Thus the theorem's content is essentially the assumption restated: 'discriminative primitives are universally superior' becomes 'local minima drive non-discriminative primitives to zero.' Because Table IV checks a different quantity (logit margins) than the pairwise condition, the assumption is not externally validated, and no independent evidence breaks the circularity. A further gap (not circularity per se) is that the analyzed model φ(x)=Σ_i a_i f_i with learnable a_i does not match the baseline's fixed uniform pooling of Eq. (3), so the theorem's mechanism may not be the mechanism in the experiments. No load-bearing self-citation or uniqueness-imported-from-authors pattern was found; reference [6] is used as a framework, not as the proof of the shortcut. Overall the central theoretical causal claim reduces to its own premise, while the method itself has independent benchmark content.

Axiom & Free-Parameter Ledger

6 free parameters · 4 axioms · 3 invented entities

What the reader 'pays for': the empirical method stands on ablations and comparisons, but its reported gains depend on per-dataset tuning of M, U, I, λ, β, and the interpretation rests on an assumption (Assumption 1) that effectively embeds its own conclusion. The compositional machinery is largely self-referential to the group's [6]. The theory contributes no measured or externally anchored constant.

free parameters (6)
  • M (# common primitive sub-sets) = 4 (CIFAR-100), 3 (miniImageNet), 2 (CUB200)
    Tuned per benchmark (Table III and Sec. V-C); controls capacity of the common set; accuracy varies by ~1 pt across M=2/4/6.
  • U (# primitives per sub-set) = 60 or 80 (CIFAR-100), 100 (miniImageNet, CUB200)
    Tuned; 40/60/80 sweep in Table III shows ~0.8 pt spread.
  • I (# discriminative primitives) = 200-280 (CIFAR-100), 300 (miniImageNet), 200 (CUB200)
    Tuned; 200/240/280 sweep in Table III.
  • λ (weight of L_sim) = not stated
    Sensitivity shown in Fig. 8; exact operating value and ranges not given in main text.
  • β (weight of L_pro) = not stated
    Sensitivity shown in Fig. 8; exact value not given.
  • activation threshold for primitive activation analysis = not specified
    Sec. III-C.1: 'a predetermined threshold' — value never given; the activation-ratio evidence depends on it.
axioms (4)
  • ad hoc to paper Assumption 1 (Non-Degradation of Competing Classes): for all i∉S, s∈S, k≠y, ⟨w_k−w_y, f_s−f_i⟩ ≤ 0
    Sec. III-C.3. The theorem's conclusion (concentration on S) is a direct corollary; the condition is asserted, and its 'verification' (Table IV) measures logit margins, a different quantity.
  • ad hoc to paper Representation as weighted sum of learnable nonnegative primitive weights: φ(x)=Σ a_i f_i
    Sec. III-C.3. The gradient mechanism over a_i is the basis of Theorem 1, but the experimental baseline (ResNet-12/ViT with pooling) does not train per-primitive attention weights; the mapping is informal.
  • domain assumption Compositional view: image patches are 'visual primitives' that are decomposed and recombined across sessions
    Sec. III-B, following [6] (same group). The interpretation of Eq. 3 as evidence of primitive-level learning is a conceptual lens, not a derived fact.
  • standard math Standard calculus facts (chain rule, softmax-CE gradient)
    Lemmas 1-2; fine.
invented entities (3)
  • Regional shortcut no independent evidence
    purpose: Name for the hypothesized tendency of base training to concentrate on maximally discriminative primitives, causing transferred bias
    The entity is operationalized only through this paper's own heatmaps and activation-ratio measurements; no externally checkable prediction (e.g., a novel architecture-independent invariant) is offered.
  • Common primitive set R_u (with latent basic-level sub-sets) no independent evidence
    purpose: Explicit shared/transferable feature bank used for both base and novel classification
    Instrumental component of the method; exists only inside ARS-CDP.
  • Discriminative primitive set R_i no independent evidence
    purpose: Feature bank to preserve base-class separability
    Instrumental method component; no evidence outside the paper's ablations.

pith-pipeline@v1.3.0-alltime-deepseek · 19407 in / 24968 out tokens · 239642 ms · 2026-08-01T05:53:37.606028+00:00 · methodology

0 comments
Cite this review

Pith. "Pith review of Alleviating Regional Shortcuts for Few-Shot Class-Incremental Learning." pith.science (2026). https://pith.science/paper/M6Q3BRWM

@misc{pith2026260722072,
  author       = {Pith},
  title        = {Pith review of: Alleviating Regional Shortcuts for Few-Shot Class-Incremental Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/M6Q3BRWM}},
  note         = {Machine review of arXiv:2607.22072}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Few-shot class-incremental learning (FSCIL) aims to incrementally learn novel classes with only a few samples while avoiding forgetting base classes. However, current methods show a tendency to misclassify novel-class samples into base classes, which we find to be caused by the excessive focus on base-class-discriminative regions on novel-class samples. In this work, we aim to explore the underlying mechanism for an interpretation and solution. We first provide a compositional view to analyze the transferred and reused spatial patterns on novel-class samples. Then, through extensive experiments and theoretical analysis, we identify both empirically and theoretically that a shortcut exists in the model's base-class training, which naturally forms the excessive focus on only the most discriminative regions (primitives), which we term as the regional shortcut. Finally, based on this interpretation, to address this problem, we propose a compositional-learning-based method to learn two primitive sets (a common set and a discriminative set), which alleviates the regional shortcut by constraining the model to learn and utilize the common primitive set for base- and novel-class recognition. Extensive experiments on standard FSCIL benchmarks demonstrate the effectiveness of our approach, yielding consistent improvements over existing state-of-the-art methods in both accuracy and interpretability.

Figures

Figures reproduced from arXiv: 2607.22072 by Haichen Zhou, Ruixuan Li, Yazhe Lyu, Yixiong Zou, Yuhua Li.

Figure 1
Figure 1. Figure 1: (a) FSCIL models tend to make wrong predictions on novel-class samples (pred [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: More visualization of misclassified samples and transferred primitives for re-composition. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: (a) Activated primitives to show the re-composition on novel classes, where the baseline method can only activate (focus on) small regions that are [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: (a) The baseline model tends to only focus on the most discriminative regions/primitives for base classes, which we term as [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Our method comprises two main stages: the primitive set building stage and the primitive set utilization stage. (a) The primitive set building [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Comparison during each incremental session. [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Visualization shows that the common feature map [PITH_FULL_IMAGE:figures/full_fig_p009_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Influence of hyper-parameters rules out data augmentation as the cause [PITH_FULL_IMAGE:figures/full_fig_p009_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: (a) Comparison of distributions reveals that our method enhances the [PITH_FULL_IMAGE:figures/full_fig_p010_9.png] view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

44 extracted references · 6 linked inside Pith

  1. [1]

    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,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 139–149, 2022

    2022

  2. [2]

    Few-shot class-incremental learning,

    X. Tao, X. Hong, X. Chang, S. Dong, X. Wei, and Y . Gong, “Few-shot class-incremental learning,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 12183–12192, 2020

    2020

  3. [3]

    Distilling the knowledge in a neural network,

    G. Hinton, O. Vinyals, and J. Dean, “Distilling the knowledge in a neural network,”arXiv preprint arXiv:1503.02531, 2015

    Pith/arXiv arXiv 2015

  4. [4]

    Few-shot in- cremental learning with continually evolved classifiers,

    C. Zhang, N. Song, G. Lin, Y . Zheng, P. Pan, and Y . Xu, “Few-shot in- cremental learning with continually evolved classifiers,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 12455–12464, 2021

    2021

  5. [5]

    Few- shot class-incremental learning via training-free prototype calibration,

    Q.-W. Wang, D.-W. Zhou, Y .-K. Zhang, D.-C. Zhan, and H.-J. Ye, “Few- shot class-incremental learning via training-free prototype calibration,” Advances in Neural Information Processing Systems, vol. 36, pp. 15060– 15076, 2023

    2023

  6. [6]

    Compositional few- shot class-incremental learning,

    Y . Zou, S. Zhang, H. Zhou, Y . Li, and R. Li, “Compositional few- shot class-incremental learning,” inProceedings of the 41st International Conference on Machine Learning, pp. 62964–62977, 2024

    2024

  7. [7]

    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, pp. 831–839, 2019

    2019

  8. [8]

    Margin-based few-shot class- incremental learning with class-level overfitting mitigation,

    Y . Zou, S. Zhang, Y . Li, and R. Li, “Margin-based few-shot class- incremental learning with class-level overfitting mitigation,”Advances in neural information processing systems, vol. 35, pp. 27267–27279, 2022

    2022

  9. [9]

    Catastrophic forgetting in connectionist networks,

    R. M. French, “Catastrophic forgetting in connectionist networks,” Trends in cognitive sciences, vol. 3, no. 4, pp. 128–135, 1999

    1999

  10. [10]

    Self-promoted prototype refinement for few-shot class-incremental learning,

    K. Zhu, Y . Cao, W. Zhai, J. Cheng, and Z.-J. Zha, “Self-promoted prototype refinement for few-shot class-incremental learning,” inPro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 6801–6810, 2021

    2021

  11. [11]

    Delve into base- novel confusion: Redundancy exploration for few-shot class-incremental learning,

    H. Zhou, Y . Zou, R. Li, Y . Li, and K. Xiao, “Delve into base- novel confusion: Redundancy exploration for few-shot class-incremental learning,”arXiv preprint arXiv:2405.04918, 2024

    Pith/arXiv arXiv 2024

  12. [12]

    Few-shot class-incremental learning via generative co-memory regularization,

    K. Bao, Y . Li, D. Zeng, and S. Ge, “Few-shot class-incremental learning via generative co-memory regularization,”arXiv preprint arXiv:2601.07117, 2026

    arXiv 2026

  13. [13]

    Pki: Prior knowledge-infused neural network for few-shot class-incremental learn- ing,

    K. Bao, F. Lin, Z. Wang, Y . Li, D. Zeng, and S. Ge, “Pki: Prior knowledge-infused neural network for few-shot class-incremental learn- ing,”Neural Networks, p. 107724, 2025

    2025

  14. [14]

    Cd2: constrained dataset distillation for few-shot class-incremental learning,

    K. Bao, D. Zhang, H. Zhang, Y . Li, Y . Yue, and S. Ge, “Cd2: constrained dataset distillation for few-shot class-incremental learning,” inProceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence, pp. 4761–4769, 2025

    2025

  15. [15]

    Divide and conquer: Static-dynamic collaboration for few-shot class-incremental learning,

    K. Bao, D. Zhang, Y . Li, D. Zeng, and S. Ge, “Divide and conquer: Static-dynamic collaboration for few-shot class-incremental learning,” inProceedings of the 2025 International Conference on Multimedia Retrieval, pp. 5–14, 2025

    2025

  16. [16]

    Attraction diminishing and distributing for few-shot class-incremental learning,

    L.-J. Zhao, Z.-D. Chen, Y . Wang, X. Luo, and X.-S. Xu, “Attraction diminishing and distributing for few-shot class-incremental learning,” in Proceedings of the Computer Vision and Pattern Recognition Confer- ence, pp. 25657–25666, 2025

    2025

  17. [17]

    Language-inspired relation transfer for few-shot class-incremental learning,

    Y . Zhao, J. Li, Z. Song, and Y . Tian, “Language-inspired relation transfer for few-shot class-incremental learning,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 47, no. 2, pp. 1089–1102, 2024

    2024

  18. [18]

    Recognition-by-components: a theory of human image understanding.,

    I. Biederman, “Recognition-by-components: a theory of human image understanding.,”Psychological review, vol. 94, no. 2, p. 115, 1987

    1987

  19. [19]

    Context-based and diversity-driven specificity in compositional zero-shot learning,

    Y . Li, Z. Liu, H. Chen, and L. Yao, “Context-based and diversity-driven specificity in compositional zero-shot learning,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 17037–17046, 2024. JOURNAL OF LATEX CLASS FILES, VOL. 18, NO. 9, SEPTEMBER 2020 12

    2024

  20. [20]

    Prompting language- informed distribution for compositional zero-shot learning,

    W. Bao, L. Chen, H. Huang, and Y . Kong, “Prompting language- informed distribution for compositional zero-shot learning,” inEuropean Conference on Computer Vision, pp. 107–123, Springer, 2024

    2024

  21. [21]

    Decompose novel into known: Part concept learning for 3d novel class discovery,

    T. Weng, J. Xiao, and H. Jiang, “Decompose novel into known: Part concept learning for 3d novel class discovery,”Advances in Neural Information Processing Systems, vol. 36, pp. 54282–54299, 2023

    2023

  22. [22]

    Disentangled representation learning,

    X. Wang, H. Chen, S. Tang, Z. Wu, and W. Zhu, “Disentangled representation learning,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 46, no. 12, pp. 9677–9696, 2024

    2024

  23. [23]

    Layer-wise representation fusion for compositional generaliza- tion,

    Y . Zheng, L. Lin, S. Li, Y . Yuan, Z. Lai, S. Liu, B. Fu, Y . Chen, and X. Shi, “Layer-wise representation fusion for compositional generaliza- tion,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 38, pp. 19706–19714, 2024

    2024

  24. [24]

    Learning clustering- based prototypes for compositional zero-shot learning,

    H. Qu, J. Wei, X. Shu, and W. Wang, “Learning clustering- based prototypes for compositional zero-shot learning,”arXiv preprint arXiv:2502.06501, 2025

    Pith/arXiv arXiv 2025

  25. [25]

    Parts of recognition,

    D. D. Hoffman and W. A. Richards, “Parts of recognition,”Cognition, vol. 18, no. 1-3, pp. 65–96, 1984

    1984

  26. [26]

    Learning deep features for discriminative localization,

    B. Zhou, A. Khosla, A. Lapedriza, A. Oliva, and A. Torralba, “Learning deep features for discriminative localization,” inProceedings of the IEEE conference on computer vision and pattern recognition, pp. 2921–2929, 2016

    2016

  27. [27]

    Compo- sitional few-shot recognition with primitive discovery and enhancing,

    Y . Zou, S. Zhang, K. Chen, Y . Tian, Y . Wang, and J. M. Moura, “Compo- sitional few-shot recognition with primitive discovery and enhancing,” in Proceedings of the 28th ACM International Conference on Multimedia, pp. 156–164, 2020

    2020

  28. [28]

    Basic objects in natural categories,

    E. Rosch, C. B. Mervis, W. D. Gray, D. M. Johnson, and P. Boyes- Braem, “Basic objects in natural categories,”Cognitive psychology, vol. 8, no. 3, pp. 382–439, 1976

    1976

  29. [29]

    Learning orthogonal prototypes for generalized few-shot semantic segmentation,

    S.-A. Liu, Y . Zhang, Z. Qiu, H. Xie, Y . Zhang, and T. Yao, “Learning orthogonal prototypes for generalized few-shot semantic segmentation,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 11319–11328, 2023

    2023

  30. [30]

    Learning deep representations by mutual information estimation and maximization,

    R. D. Hjelm, A. Fedorov, S. Lavoie-Marchildon, K. Grewal, P. Bach- man, A. Trischler, and Y . Bengio, “Learning deep representations by mutual information estimation and maximization,”arXiv preprint arXiv:1808.06670, 2018

    Pith/arXiv arXiv 2018

  31. [31]

    Neural collapse inspired feature-classifier alignment for few-shot class incremental learn- ing,

    Y . Yang, H. Yuan, X. Li, Z. Lin, P. Torr, and D. Tao, “Neural collapse inspired feature-classifier alignment for few-shot class incremental learn- ing,”arXiv preprint arXiv:2302.03004, 2023

    Pith/arXiv arXiv 2023

  32. [32]

    Few- shot class-incremental learning by sampling multi-phase tasks,

    D.-W. Zhou, H.-J. Ye, L. Ma, D. Xie, S. Pu, and D.-C. Zhan, “Few- shot class-incremental learning by sampling multi-phase tasks,”IEEE Transactions on Pattern Analysis and Machine Intelligence, 2022

    2022

  33. [33]

    Metafscil: a meta-learning approach for few-shot class incremental learning,

    Z. Chi, L. Gu, H. Liu, Y . Wang, Y . Yu, and J. Tang, “Metafscil: a meta-learning approach for few-shot class incremental learning,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 14166–14175, 2022

    2022

  34. [34]

    For- ward compatible few-shot class-incremental learning,

    D.-W. Zhou, F.-Y . Wang, H.-J. Ye, L. Ma, S. Pu, and D.-C. Zhan, “For- ward compatible few-shot class-incremental learning,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 9046–9056, 2022

    2022

  35. [35]

    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,” inComputer Vision–ECCV 2022: 17th European Conference, Tel Aviv, Israel, October 23–27, 2022, Proceedings, Part XXIV, pp. 146–162, Springer, 2022

    2022

  36. [36]

    Few-shot class- incremental learning from an open-set perspective,

    C. Peng, K. Zhao, T. Wang, M. Li, and B. C. Lovell, “Few-shot class- incremental learning from an open-set perspective,” inComputer Vision– ECCV 2022: 17th European Conference, Tel Aviv, Israel, October 23–27, 2022, Proceedings, Part XXV, pp. 382–397, Springer, 2022

    2022

  37. [37]

    Rethinking few- shot class-incremental learning: Learning from yourself,

    Y .-M. Tang, Y .-X. Peng, J. Meng, and W.-S. Zheng, “Rethinking few- shot class-incremental learning: Learning from yourself,”arXiv preprint arXiv:2407.07468, 2024

    Pith/arXiv arXiv 2024

  38. [38]

    Pre-trained vision and language transformers are few-shot incremental learners,

    K.-H. Park, K. Song, and G.-M. Park, “Pre-trained vision and language transformers are few-shot incremental learners,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 23881–23890, 2024

    2024

  39. [39]

    Prototype-guided memory replay for continual learning,

    S. Ho, M. Liu, L. Du, L. Gao, and Y . Xiang, “Prototype-guided memory replay for continual learning,”IEEE transactions on neural networks and learning systems, vol. 35, no. 8, pp. 10973–10983, 2023

    2023

  40. [40]

    Learning multiple layers of features from tiny images,

    A. Krizhevsky, G. Hinton,et al., “Learning multiple layers of features from tiny images,” 2009

    2009

  41. [41]

    Imagenet large scale visual recognition challenge,

    O. Russakovsky, J. Deng, H. Su, J. Krause, S. Satheesh, S. Ma, Z. Huang, A. Karpathy, A. Khosla, M. Bernstein,et al., “Imagenet large scale visual recognition challenge,”International journal of computer vision, vol. 115, pp. 211–252, 2015

    2015

  42. [42]

    The caltech-ucsd birds-200-2011 dataset,

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

    2011

  43. [43]

    Visualizing data using t-sne.,

    L. Van der Maaten and G. Hinton, “Visualizing data using t-sne.,” Journal of machine learning research, vol. 9, no. 11, 2008

    2008

  44. [44]

    Learning with fantasy: Semantic-aware virtual contrastive constraint for few-shot class- incremental learning,

    Z. Song, Y . Zhao, Y . Shi, P. Peng, L. Yuan, and Y . Tian, “Learning with fantasy: Semantic-aware virtual contrastive constraint for few-shot class- incremental learning,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 24183–24192, 2023. VI. BIOGRAPHYSECTION Haichen Zhoureceived the BS degree from the School of M...

    2023