REVIEW 3 cited by
Adversarially Robust Generalization Just Requires More Unlabeled Data
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
read the original abstract
Neural network robustness has recently been highlighted by the existence of adversarial examples. Many previous works show that the learned networks do not perform well on perturbed test data, and significantly more labeled data is required to achieve adversarially robust generalization. In this paper, we theoretically and empirically show that with just more unlabeled data, we can learn a model with better adversarially robust generalization. The key insight of our results is based on a risk decomposition theorem, in which the expected robust risk is separated into two parts: the stability part which measures the prediction stability in the presence of perturbations, and the accuracy part which evaluates the standard classification accuracy. As the stability part does not depend on any label information, we can optimize this part using unlabeled data. We further prove that for a specific Gaussian mixture problem, adversarially robust generalization can be almost as easy as the standard generalization in supervised learning if a sufficiently large amount of unlabeled data is provided. Inspired by the theoretical findings, we further show that a practical adversarial training algorithm that leverages unlabeled data can improve adversarial robust generalization on MNIST and Cifar-10.
Forward citations
Cited by 3 Pith papers
-
On the existence of consistent adversarial attacks in high-dimensional linear classification
The authors derive sharp high-dimensional formulas for consistent adversarial errors in linear classifiers and show overparameterization increases vulnerability on correctly classified points while decreasing the over...
-
Adversarial Training from Mean Field Perspective
A mean field framework for random ReLU networks yields adversarial-loss bounds and predicts that adversarial training shrinks weights, hurts vanilla depth, and is rescued by residual connections and width.
-
Understanding Adversarial Training with Energy-based Models
Delta energy, the energy gap between an image and its adversarial counterpart, separates catastrophic from robust overfitting, and penalizing it with the DER regularizer mitigates both while improving generation diversity.
Discussion (0). Sign in to comment.