- Rating
- 2
- Confidence
- 2
- Informativeness
- 2
- Presentation
- 2
- Soundness
- 2
Read review
Paper summary
This paper addresses adversarial robustness in Few-Shot Class-Incremental Learning under severe data scarcity. It proposes FSCIL-AT, combining session-wise LoRA experts, adversarial two-player game optimization, and CKA regularization. Experiments on four benchmarks evaluate clean and robust accuracy under multiple attacks, with ablations and few-shot robustness analyses.
Critical strengths
The paper investigates an important and relatively underexplored problem by introducing adversarial robustness into the FSCIL setting.
Critical weaknesses
The experimental evidence is insufficient to fully support the paper’s main claims. The paper claims to address catastrophic forgetting, data scarcity, and adversarial fragility, which are major challenges in FSCIL and adversarial robustness. These broad claims require more extensive and direct experimental evidence.
Questions for authors
- The overall story, method, and experiments are not yet fully aligned. The paper spans FSCIL and adversarial robustness and claims to address catastrophic forgetting, data scarcity, and adversarial fragility. However, the current experimental results mainly focus on adversarial robustness and do not sufficiently support all of these claims. The authors should reconsider the core problem and main focus of the method, and ensure that the motivation, method design, and experiments form a clear and consistent logical chain.
- The experimental evidence is unbalanced with respect to the three challenges claimed in the abstract. While adversarial fragility and data scarcity are explicitly studied, the claim of mitigating catastrophic forgetting is only indirectly supported by session-averaged clean accuracy and performance-drop curves. More direct continual-learning metrics, such as old/new-class accuracy would be necessary to substantiate this claim.
- Since the central motivation is robustness under extreme data scarcity, the robustness analysis should be more systematic. Fig. 1 varies the number of shots on CUB-200, which is useful, but this evidence is limited to one benchmark and does not examine other important FSCIL factors, such as the number of incremental classes, session length, or class-growth rate. A broader robustness analysis across different scarcity would better support the claimed generality of the method.
- The theoretical analysis appears to be inconsistent with the actual optimization of the proposed method. In the theoretical analysis, the attacker and defender are formulated using a shared minimax objective, while in the actual training process, the defender is optimized with a joint loss containing the clean loss, transferred adversarial loss, self-generated adversarial loss, and CKA regularization. Therefore, the optimization analyzed in the theory is not fully identical to the implemented training objective. In addition, the strong convex-concavity assumption required for the claimed linear convergence is not verified for the deep LoRA model with adversarial inner optimization. The authors should reconsider whether the current theoretical analysis can rigorously support the proposed method.
- The problem positioning appears broader than the actual technical contribution. The abstract frames the work as simultaneously addressing catastrophic forgetting, data scarcity, and adversarial fragility, whereas the central technical novelty and most experimental evidence focus on adversarial robustness with data scarcity. The FSCIL mechanisms for retaining previous knowledge are comparatively conventional and are not independently validated to the same extent. The paper would benefit from a more focused problem statement clarifying whether its primary goal is robust FSCIL or, more specifically, robust adaptation under extreme few-shot supervision.
- Although the paper provides module-level ablations for LWG, ATPGO, and MBCKAR, the contribution of the individual terms within ATPGO remains insufficiently isolated. In particular, the defender combines clean, transferred-adversarial, self-adversarial, and CKA losses. Controlled comparisons such as self-only, transfer-only, self+transfer, and static-versus-co-evolving attacker variants would better establish whether the proposed co-evolution mechanism itself, rather than simply additional adversarial supervision, is responsible for the improvement.
- The continual robustness and scalability of the method are not sufficiently evaluated. The paper mainly reports robust performance after the final session, while it remains unclear how the robustness of old and new classes evolves as more sessions are introduced. Moreover, since a new LoRA expert is created for each session and ATPGO and MBCKAR are repeatedly performed, both computational and storage costs may grow with the number of sessions. The authors should evaluate the robustness evolution across sessions and analyze the training, inference, and storage costs under longer continual-learning sequences.
- The main paper is not sufficiently self-contained. Key experiments and experimental settings supporting the claims on catastrophic forgetting, data scarcity, and adversarial fragility are not fully presented in the main text, with important details deferred to the appendix. The authors should include the essential evidence and settings in the main paper, since reviewers are not required to read the supplementary material.
9 comments
Response 2/3 to Reviewer ecV2
Q4: Problem Scope and Contribution Focus
A4: We thank the reviewer for encouraging this clarification. We clarify the issue below.
(1) Unified Regularized Minimax Objective.
In Eq.(14), denotes the attacker-defender coupling term rather than the complete defender objective. Specifically, define
$R_t(\Theta_t)=\mathbb{E}_{(\mathbf{x},y)\sim\mathcal{D}_t}\left[\mathcal{L}_{\mathrm{CE}}(\mathcal{E}_t,\mathbf{x},y)+ \lambda_2\mathcal{L}_{\mathrm{CE}}(\mathcal{E}_t,\mathbf{x}^{\mathrm{self}},y)\right]+\lambda_{\mathrm{CKA}}\mathcal{L}_{\mathrm{CKA}},$ where the CKA term is omitted for the base session, and . Because is independent of , $\nabla_{\widetilde{\Theta}_t}\bar{u}_t= \lambda_1\nabla_{\widetilde{\Theta}_t}u$ , $\nabla_{\Theta_t}\bar{u}_t=\lambda_1\nabla_{\Theta_t}u+\nabla_{\Theta_t}R_t= \nabla_{\Theta_t}\mathcal{L}_{\mathrm{def}}$. For , the scaling of the attacker gradient is absorbed into its step size and does not change the ascent direction. Thus, the implemented two-stage procedure corresponds to gradient ascent/descent on the augmented minimax payoff . The clean, self-generated, and CKA terms are defender-side regularizers within the same game, rather than objectives external to the theoretical formulation. Eq.(14) isolates the coupled transfer term, whereas Eq.(13) specifies the complete regularized implementation.
(2) Conditional Linear Convergence.
The linear-convergence proposition characterizes the effective ATPGO game operator after replacing the coupling payoff with the augmented payoff , under local smoothness and strong convexity-concavity of . Under these conditions, the corresponding operator is strongly monotone and Lipschitz continuous, yielding a linear contraction toward the local Nash equilibrium. The theorem concerns the actual alternating ATPGO updates in the LoRA parameter space and does not require the original deep-network loss to be globally convex. The frozen backbone, low-dimensional LoRA adaptation, and local treatment of the adversarial generation map define the effective game analyzed by the proposition. Hence, deep-model nonlinearity does not invalidate the result; the stated guarantee is a local structural characterization rather than an unconditional global-convergence claim. This conditional treatment is consistent with established minimax analyses under explicit structural assumptions[1][2].
Q5: Problem Scope and Contribution Focus
A5: We thank the reviewer for this valuable suggestion. In the revised manuscript, we adjusted the motivation and problem statement to clarify that our primary goal is robust adaptation under extreme few-shot continual supervision. Our method addresses limited samples and sparse class spaces in each session by balancing clean accuracy, adversarial robustness, and continual knowledge retention.
Q6: ATPGO Mechanism-Level Validation
A6: We thank the reviewer for this helpful suggestion. We conduct ablations anlysis, results in Tab.R6 and R7. ''Clean + S/T'' denotes clean training with self-generated / transferred supervision, ''Clean + T + S'' combines both, and ''Frozen'' versus ''Co-evolving'' indicates whether the attacker is fixed or jointly updated with the defender.
Tab. R6: Ablations of ATPGO components and attacker co-evolution in CIFAR100.
| Setting | Attacker | Clean Avg. | ℓ∞-PGD | ℓ2-PGD | FGSM | BIM | CW |
|---|---|---|---|---|---|---|---|
| Clean + S | Co-evolving | 79.75 | 7.02 | 33.73 | 46.76 | 5.61 | 36.50 |
| Clean + T | Co-evolving | 81.07 | 6.81 | 33.58 | 49.68 | 4.40 | 36.85 |
| Clean + T + S | Frozen attacker | 80.06 | 9.80 | 35.72 | 50.06 | 7.43 | 37.84 |
| Clean + T + S | Co-evolving | 81.96 | 11.71 | 44.11 | 52.12 | 8.94 | 38.65 |
Tab. R7: Ablations of ATPGO components and attacker co-evolution in CUB200.
| Setting | Attacker | Clean Avg. | ℓ∞-PGD | ℓ2-PGD | FGSM | BIM | CW |
|---|---|---|---|---|---|---|---|
| Clean + S | Co-evolving | 76.83 | 1.12 | 47.85 | 30.61 | 1.10 | 43.02 |
| Clean + T | Co-evolving | 76.42 | 1.42 | 43.91 | 29.93 | 1.44 | 41.64 |
| Clean + T + S | Frozen attacker | 78.91 | 1.73 | 49.56 | 31.33 | 1.62 | 46.34 |
| Clean + T + S | Co-evolving | 81.41 | 2.42 | 52.82 | 32.35 | 2.94 | 50.54 |
The co-evolving model performs best on both datasets, showing gains beyond additional adversarial supervision.
[1] Finding Second-Order Stationary Points in Nonconvex-Strongly-Concave Minimax Optimization, NeurIPS 2022
[2] Efficient Algorithms for Smooth Minimax Optimization, NeurIPS 2019
Response 3/3 to Reviewer ecV2
Q7: Continual Robustness and Computational Scalability
A7: We thank the reviewer for this suggestion and evaluate session-wise robustness and computational costs below.
(1) Session-wise Robustness.
Tab. R8 reports old/new robust accuracy (%) averaged over FGSM, -PGD, and CW, with entries ordered as old / new. -PGD and BIM are omitted due to uniformly low accuracies and limited inter-method discrimination. FSCIL-Adv achieves the strongest overall robustness.
Tab. R8: Session-wise adversarial robustness on CUB-200 (%).
| SESSION | FSCIL-Adv | CL-LoRA | KeepLoRA | ARMCL |
|---|---|---|---|---|
| S0 | — / 60.89 | — / 58.76 | — / 51.44 | — / 32.77 |
| S1 | 54.68 / 56.23 | 57.98 / 35.01 | 50.73 / 22.89 | 29.38 / 16.16 |
| S2 | 53.96 / 26.29 | 55.07 / 16.53 | 48.27 / 11.65 | 21.05 / 8.94 |
| S3 | 51.40 / 19.94 | 50.76 / 16.07 | 44.65 / 12.50 | 25.92 / 8.33 |
| S4 | 48.24 / 45.39 | 48.04 / 14.87 | 41.53 / 14.61 | 24.18 / 20.00 |
| S5 | 46.91 / 29.53 | 44.63 / 20.47 | 38.55 / 9.94 | 22.45 / 14.62 |
| S6 | 44.92 / 36.36 | 42.63 / 23.94 | 36.31 / 16.36 | 20.83 / 10.91 |
| S7 | 43.93 / 34.84 | 41.27 / 35.34 | 34.09 / 13.03 | 15.35 / 7.02 |
| S8 | 42.64 / 23.69 | 40.66 / 20.11 | 32.30 / 8.54 | 16.22 / 9.92 |
| S9 | 41.44 / 57.10 | 39.09 / 41.53 | 30.13 / 24.04 | 14.10 / 24.86 |
| S10 | 42.00 / 49.86 | 39.00 / 44.72 | 29.51 / 16.26 | 17.20 / 28.73 |
| Avg | 47.01 / 37.92 | 45.91 / 26.86 | 38.61 / 14.98 | 20.67 / 14.95 |
(2) Computational and Storage Costs.
We compare FSCIL-Adv with native robust baselines using PGD- for AIR/DSEF and PGD- for ARMCL. In Tab. R9, denotes the number of new classes per session and the total number of sessions. Each cell reports training time (min) / clean-forward latency (ms/image) / model tensor footprint (MiB) for W5-S21, W10-S11, and W20-S6. Overall, our method remains within the average cost range.
Tab. R9: Cost comparison on CUB-200.
| METHOD | W5-S21 | W10-S11 | W20-S6 |
|---|---|---|---|
| FSCIL-Adv | 77.38 / 44.30 / 434.69 | 66.61 / 28.06 / 383.83 | 61.48 / 14.43 / 358.40 |
| AIR-PGD10 | 48.61 / 35.74 / 398.12 | 40.78 / 17.26 / 364.73 | 33.45 / 6.84 / 347.66 |
| DSEF-PGD10 | 92.94 / 46.88 / 410.36 | 82.17 / 28.33 / 371.92 | 72.80 / 17.21 / 350.88 |
| ARMCL-PGD20 | 83.47 / 47.15 / 417.48 | 67.92 / 28.60 / 378.14 | 64.61 / 17.38 / 353.72 |
Q8: Self-Contained Experimental Evidence
A8: We thank the reviewer for this valuable structural suggestion. In the revised manuscript, we moved the essential experimental settings from the Appendix C to the main text, including class splits and shots, backbone and training schedules, attack budgets and iterations. We also incorporated evaluations under a broader range of attack scenarios, analyses of forgetting rates, comparisons across different few-shot settings, and attack-wise component ablations from Appendix F, providing more comprehensive empirical support for our core claims.
Response 1/3 to Reviewer ecV2
We sincerely thank the reviewer for the constructive feedback. We have strengthened the experimental evidence and refined the relevant content accordingly, as detailed below.
Q1: Alignment of Motivation, Method, and Experimental Evidence
A1: We thank the reviewer for the valuable suggestion to strengthen the paper’s overall logical consistency. We will adjust the problem statement in the abstract and introduction to emphasize robust adaptation under extreme few-shot class-incremental supervision in the revised version. ATPGO addresses adversarial fragility under limited samples and sparse per-session supervision, while LWG and MBCKAR support knowledge retention and feature complementarity. Building on the reviewer’s subsequent comments, we further provide comprehensive experimental evidence supporting our method, including old/new-class accuracy, forgetting rate, and multifactor analyses covering shots, base-class size, session length, and class-growth rate. These experiments and analyses are also included in the Appendix to strengthen the paper’s overall coherence.
Q2: Direct Evaluation of Continual Forgetting
A2: We thank the reviewer for this valuable comment. Tab. R1 report per-session old / new-class accuracy and forgetting rate. old / new accuracy denotes mean class accuracy over previously learned / current classes, while forgetting rate measures the relative drop from each old class’s historical best accuracy.
Tab. R1: Old/new-class accuracy (%) and forgetting rate (%) on CUB-200.
FSCIL-Adv achieves the strongest overall results, indicating better retention of previously learned classes while maintaining effective adaptation to new classes.
Q3: Robustness Generalization across FSCIL Factors
A3: We thank the reviewer for this valuable suggestion. We additionally report the same few-shot analysis on CIFAR-100 in Tab. R2, including rival exposure / robust generalization gap / held-out test robustness. ATPGO consistently achieves broader rival exposure, a smaller gap, and higher held-out robustness.
Tab. R2: Few-shot adversarial robustness and rival exposure on CIFAR-100.
We further extend the analysis to three FSCIL factors in Tabs. R3–R5.
Tab. R3: Base-class size and session length on CUB200
Tab. R4: Data scarcity by shots per class on CUB200
Tab. R5: Class-growth rate and session length on CUB200
Here, B, W, K, and S denote base classes, new classes per session, shots per class, and total sessions, respectively. Each entry reports Clean / Rob-Avg (%), where Rob-Avg averages five attacks. The original protocol is B100-W10-K5-S11. Tab. R3 varies B, Tab. R4 varies K, and Tab. R5 fixes the final class count and K=5 while varying W, thereby changing session length and the growth rate W/B. FSCIL-AT achieves the highest Rob-Avg in every setting and the strongest clean accuracy in nearly all configurations, demonstrating robustness to diverse data-scarcity and class-growth conditions.