Adversarial examples and data concentration
The bibliography cites a paper on how the concentration of data affects adversarial robustness. Its title asks whether adversarial examples can be avoided, which puts the data distribution at the centre of the question.
The requirement in plain English
The bibliography cites a paper on how the concentration of data affects adversarial robustness. Its title asks whether adversarial examples can be avoided, which puts the data distribution at the centre of the question.
What the public record says
The bibliography entry names the authors, title and venue. The public view gives no summary and no clause reference.
Why it matters to you
This citation is a warning against treating robustness as a property of the model alone. If the data concentrate, the effective input region is smaller than the nominal space, and a robustness result may say more about the data than about the network. For a buyer, the practical consequence is that the dataset description is part of the evidence, not an appendix. A result that cannot be tied to a data distribution cannot be compared with another result.
Evidence a reviewer asks for
- The data distribution description, including the concentration or clustering found.
- The relationship between the test distribution and the deployment distribution.
- The adversarial search method, budget and success rate.
- A note on whether the result is expected to transfer.
Checks to run this month
- Does the evidence describe the data distribution, not only the dataset name?
- Is the test distribution representative of deployment?
- Would the result transfer to a new source?
Dates
Paper published 2023. FDIS registered 2026-08-21.
Cross-references
- High-dimensional input and why exhaustive testing fails
- Corruption robustness and its efficient assessment
- Write the sampling and perturbation plan
What this page does not state
The public view gives no summary and no statement of how the draft uses the paper.