- Rating
- 2
- Confidence
- 2
- Informativeness
- 2
- Presentation
- 2
- Soundness
- 2
Read review
Paper summary
Despite the strong threats from data-poisoning backdoors, their behaviors are actually counter-intuitive with varying the trigger strength. Inspired by these, the author analyze the backdoor's behaviors and show that, with a fixed test trigger, stronger trigger strength can improve the clean accuracy and meanwhile reduce backdoor's attack success rate beyond an optimal strength. The proposed high-dimensional theory also identifies low-variance data directions as the most effective trigger directions, with experiments supporting these findings.
Critical strengths
- Insightful findings of backdoor attacks
The impact of the backdoor trigger strength on clean accuracy and attack success rate is very interesting. The increasing clean accuracy and a peak attack success rate also unveil a certain entanglement and disentanglement between the natural performance and the backdoor.
- The theoretical analysis on the three high-dimensionalities is rigorous.
The extensive analysis of the three impact factors of backdoor attacks provides a strong theoretical foundation for the considered attacking situation.
Critical weaknesses
- Unclear definition of trigger strength
The trigger strength depends on multiple factors, including the trigger opacity, the position on the input, the dissimilarity to the benign samples, the poisoning ratio, etc. Particularly in Section 3, the measurement of trigger strength is unclear.
- Discussion of the theory for the application
In the abstract begin, the authors claim that the backdoor's counter-intuitive behaviors can help the defender. In the paper, however, there is no discussion about the application possibility of those interesting findings on defense methods.
- Limited consideration of attack variants.
Section D offers the poisoning setting, where a simplistic patch trigger is applied in this work. Given various existing poisoning methods, e.g., Blend [1] and WaNet [2], the adaptability of the proposed high-dimensional theory is not evaluated.
- Overly low clean accuracy of multiclass CIFAR10 with ResNet18.
In Figure 6, the clean accuracy with trigger norm from to remains lower than 90%. However, the standard training of a ResNet18 on CIFAR10 is normally at 94% ~ 95%. This lower clean accuracy requires further interpretation.
Questions for authors
-
According to the visualization in Figure 1, the trigger strength with the highest attack success rate does not achieve a high enough clean accuracy. Can the authors explain what causes this mismatch?
-
As shown in the evaluation, the logic regression models exhibits a clear alignment with the theoretical analysis in Section 3. In contrast, as mentioned by the authors, a ResNet18 exhibits similar qualitative behaviour beyond the convex setting. Can the authors interpret this problem from the viewpoint of the model capacity difference?
Response to reviewer 1Jwv
We thank the reviewer for their review and are glad they found the findings insightful.
Trigger strength. α is the ℓ2 norm of the additive perturbation αv, with ∥v∥=1 (Sec. 2, App. D). Position and shape are encoded in the direction v, and the poisoning ratio is the separate parameter ϕ. For our patch, v is uniform on 12 entries (2×2×3). So α=0.5 adds ≈0.14 and α=1.5 adds ≈0.43 per entry, in [0,1] pixel units. We have restated this at the start of Sec. 3.
Defences. We agree with the reviewer that this deserves discussion. By "helps the defender" in the abstract, we mean that past the peak, clean accuracy is higher and ASR is lower; we now say this explicitly. We have also added a short paragraph on two implications:
Attack variants. A WaNet-style warp is in App. D (Fig. 7). Clean accuracy rises from 0.949 to 0.973, and ASR peaks at 0.76 near β=3.8 then falls to 0.08. We now point to it from the introduction.
Multiclass accuracy. Our protocol is not standard CIFAR-10 training:
We now state this in App. D.
Q1. This is the trade-off itself. For population squared loss in the eigenvector setting of Cor. 4, both metrics depend on α only through r=hv/hμ: clean accuracy falls as r grows while ASR rises. In that model the ASR-optimal strength is exactly where clean accuracy is lowest. With finite samples the noise floor also affects the full margin variance (Sec. 4), but the tension remains.
Q2. Our theory concerns fixed-representation GLMs. ResNet learns its representation, so capacity and optimisation may affect both trigger features and sensitivity. Its qualitative agreement is empirical evidence beyond the theory, not a guarantee for every cross-entropy model with weight decay. The main difference we see between the two models is that the ResNet's ASR rises and falls much more sharply (Fig. 5).
We hope these clarifications address the reviewer's concerns.