- Rating
- 3
- Confidence
- 2
- Informativeness
- 2
- Presentation
- 2
- Soundness
- 2
Read review
Paper summary
This work studies the first-order behavior of Probabilistic Robustness (PR) with respect to model parameters, provide a new perspective for understanding the first-order optimization structure of PR. By applying the co-area formula, the authors formulate the gradient of the Non-Robust Probability (NPR) as a weighted integral over the decision boundary and approximate it using Gaussian kernels and Monte Carlo sampling. Based on this gradient estimator, they develop NPR-Aug, a training strategy designed to improve probabilistic robustness. Experiments show that aggregating multiple perturbations yields more reliable NPR descent directions and enhances robustness against stochastic perturbations.
Critical strengths
- The paper derives a boundary-integral representation of the NPR gradient using the co-area formula, providing a principled characterization of how decision-boundary contributions determine the first-order variation of PR.
- The experiments support the theoretical analysis by demonstrating the importance of aggregating boundary contributions. The proposed NPR-Aug. further shows that incorporating this information into training can improve probabilistic robustness.
Critical weaknesses
- The high-dimensional experiments are primarily conducted on CIFAR-10 with ResNet-18, without evaluations across diverse datasets or model architectures. As such, the current experimental scope is inadequate to demonstrate the generality of the proposed findings.
- Systematic ablations on perturbation sample size, gradient aggregation strategies, and NPR loss weight are missing, making it difficult to assess the contributions of these design choices.
Questions for authors
- Could the authors extend the experiments in Section 6.3 to a more challenging dataset, such as CIFAR-100 or ImageNet, and a different model architecture, such as Vision Transformer?
- A sensitivity analysis of NPR-Aug. with different perturbation sample sizes () and NPR loss weights () under identical training settings would help clarify the impact of these hyperparameters on performance.
2 comments
W2/Q2. Ablations on sampling, aggregation and loss weight
We added CIFAR-10/ResNet-18 experiments with the training schedule, optimizer and MMA hinge fixed. PR averages uniform, clipped Gaussian, clipped Laplace, sign, sparse and low-frequency noise, using identical evaluation images and perturbation draws within each comparison. The and scans use augmentation and one training run per setting, where .
-
How sensitive is NPR-Aug. to the number of perturbations? We vary with . Time is averaged over recorded training epochs.
Samples per image PR at 4× PR at 8× PR at 16× Seconds per recorded epoch 1 83.15 71.40 42.31 138.9 2 82.10 72.62 45.97 145.3 4 83.78 72.05 43.13 165.6 8 82.92 71.31 40.93 190.7 Increasing from four to eight adds about 15% to the recorded epoch time without improving PR at the displayed radii. The best setting varies with evaluation radius, while the paper's offers competitive performance at moderate sampling cost in this scan.
-
How sensitive is the result to the NPR loss weight? We vary with . The boundary-anchor term becomes , with other MMA terms unchanged. Thus, recovers MMA.
NPR weight PR at 4× PR at 8× PR at 16× 0.00 77.09 65.72 41.05 0.25 81.24 68.72 41.66 0.50 83.78 72.05 43.13 0.75 82.35 65.46 34.83 The paper's choice of gives the highest observed PR at all three radii in this scan. Increasing it to 0.75 lowers PR at by 8.30 points, supporting a balance between the anchor and sampled terms.
-
How do individual directions compare with aggregated boundary contributions? We vary the number of sampled boundary points on three CIFAR-10 MMA training seeds. These are diagnostic boundary points, distinct from the random perturbations used in training. Values are mean ± sample standard deviation across runs, after averaging over retained images within each run.
Boundary points Sampled-point cosine Anchor cosine 2 0.876 ± 0.008 0.688 ± 0.015 4 0.907 ± 0.005 0.688 ± 0.021 8 0.920 ± 0.002 0.701 ± 0.026 16 0.924 ± 0.004 0.711 ± 0.029 Sampled-point cosine compares each point's parameter gradient with the weighted aggregate of the other points. Anchor cosine compares the MMA anchor's gradient with all points. Weights are , and higher cosine means closer directional agreement. The anchor remains less aligned with the sampled aggregate than the sampled boundary points across all tested reference sizes and seeds. This is a directional diagnostic, not a direct comparison of aggregation strategies during training. Sampled points and retained images also vary with .
We will include these experiments and their settings in the revision.
W1/Q1. Additional datasets and architectures
Thank you for recognizing the theoretical characterization and the value of aggregating boundary contributions. We appreciate your suggestions for broader evaluation and systematic ablations, which help clarify the role of each design choice.
To extend Section 6.3 beyond CIFAR-10/ResNet-18, we added CIFAR-100 and VGG16-bn experiments. PR below averages uniform, clipped Gaussian, clipped Laplace, sign, sparse and low-frequency noise, with identical evaluation perturbations within each setting. Accuracies are percentages. NPR-Aug. uses 4× augmentation, where 1×=8/255.
1. CIFAR-100 with ResNet-18
The standard-training reference has 77.58 clean accuracy and PR of 29.25, 15.52 and 6.00 at 4×, 8× and 16×. The table compares robustness-training methods. MMA and NPR-Aug. report mean ± sample standard deviation over three seeds. Other rows are single runs.
MMA and NPR-Aug. achieve the highest mean PR among the evaluated methods at all three radii. Both exceed every other method by at least 4.4 points at 4× and 5.8 at 8×. EVaR has the highest clean accuracy among the defended models but lower PR under these stronger perturbations.
NPR-Aug. further improves clean accuracy and PR at 4× over MMA. The gains are 2.38 points (95% CI [1.35, 3.41]) and 1.46 points ([0.32, 2.61]), respectively. At larger radii, NPR-minus-MMA differences are small and their confidence intervals include zero.
A geometric interpretation. Anchor cosine measures alignment between the anchor gradient and a sampled aggregate of boundary contributions. On NPR-Aug. checkpoints, anchor cosine is 0.18 higher on CIFAR-100 than on CIFAR-10 (three seeds per dataset, 95% CI [0.09, 0.28]). For MMA, the difference is 0.05 with an interval of [−0.01, 0.11]. These observations are consistent with the possibility that a more representative anchor leaves less room for supplementation.
2. VGG16-bn on CIFAR-10
To test a different architecture, we use VGG16-bn, which has no residual connections, with the same training recipe and no hyperparameter retuning. Each row below is one training run. Standard training gives 94.36 clean accuracy and PR of 58.12, 38.96 and 30.07 at the three radii.
NPR-Aug. has the highest observed PR at 4× and 8× among these methods. It improves on MMA by 6.84 and 5.37 points and on PGD-AT by 3.05 and 5.99, while also exceeding both on clean accuracy. MMA and NPR-Aug. have similar observed PR at 16×. EVaR retains higher clean accuracy but lower PR.