Proposal(s) title:
- Closing the Computer Vision Evaluation Gap for CEN/CENELEC JTC21 AI Act Standardisation
Proposal(s) topic:
Artificial Intelligence
Societal, Economic or Technological Impacts:
Funding impact:
Societal, Economic or Technological Impacts (2nd Open Call 2029)
This short-term fellowship identified applicable metrics for facial recognition, deepfake detection, super resolution, image denoising, inpainting, and stitching. The economic impacts for companies building AI products utilizing the tasks are:
- reduced cost and uncertainty through a clear route to presumption of conformity with the AI Act;
- clear and common performance metrics for all systems increases user trust, reducing barriers to adoption, boosting interoperability, and increasing market access for providers.
These benefits accrue particularly to SMEs as they will not need to devote scarce resources to developing their own compliance routes.
Given the human rights impact of errors, facial recognition has particular societal impact. Adding applicable metrics into the prEN 18281 Evaluation standard means such AI systems will have to report these error rates. Evaluating deepfake detection systems has similar societal importance, given their impact on human rights and democratic processes.
- New standard development
-
Custom funding impact (2nd Open Call 2029)With this fellowship, I supported developing standards that are crucial to implementing the EU AI Act, a key EU policy objective. These standards are managed by CEN/CENELC JTC1 21 AI WG3 Engineering aspects, and these include namely prEN 18288 Taxonomy and prEN 18281.
Standards Development Organisation:
