Contact

Distribution Shift Robustness Audit

Read more

Description of the service

Distribution Shift Robustness Audit tests how reliably an AI model performs when deployment data differs from training data, especially in single-domain settings. With test domains, robustness is measured directly; without them, Wasserstein/DRO stress tests estimate plausible worst-case failures.
Expected results:

Report describing model deployment robustness risks

The methodology:

Conduct experiments following specifically developped SDG theory

Target:

Widely applicable

  • Organization: Fraunhofer IPA
  • Type: Test
  • Average duration: 2-4 weeks
  • Contact: max-lion.schumacher@ipa.fraunhofer.de

Improve your project
around production and
automation with AI technologies