Corruption robustness and its efficient assessment
The bibliography cites two works on corruption robustness: an evaluation framework and an efficient statistical assessment method. Both concern performance under input corruptions rather than worst-case adversarial changes.
The requirement in plain English
The bibliography cites two works on corruption robustness: an evaluation framework and an efficient statistical assessment method. Both concern performance under input corruptions rather than worst-case adversarial changes.
What the public record says
The bibliography entries name the titles, authors and years. The public view gives no summaries and no clause references.
Why it matters to you
Corruption robustness is often the closest to production reality: blur, noise, compression, weather, sensor drift and missing values. It is also where a statistical method earns its place, because the corruption space is large and the effect is naturally a distribution. The two citations together suggest the draft treats efficiency as part of the method, not an afterthought. For a buyer, the practical question is which corruptions are in the deployment risk case, and whether the measurement covers them.
Evidence a reviewer asks for
- The corruption catalogue, with severity levels and the reason each is in the risk case.
- The sampling method and the number of draws per corruption.
- The performance metric and its baseline on clean input.
- The efficiency record: compute, time and cost.
Checks to run this month
- Does the corruption catalogue match the deployment risks?
- Is the clean baseline recorded?
- Is the cost of the assessment known before it is committed?
Dates
Papers published 2021 and 2022. FDIS registered 2026-08-21.
Cross-references
- Adversarial examples and data concentration
- Set the acceptance threshold and the decision rule
- Write the sampling and perturbation plan
What this page does not state
The public view gives no summaries and no statement of how the draft uses the papers.