Towards a Comprehensive ML Auditing Framework: Extending the Core Criteria Catalog

Abstract

As more ML technologies enter crucial sectors like medicine and governance, maintaining reliability, responsibility, and openness via audits is increasingly important. The document introduces an enhanced machine learning audit system constructed upon the fundamental standards outlined in Schwarz's research group. The statement has been restated in another manner without altering its core message: [2]. A preliminary investigation into "auditable artificial intelligence" was undertaken through systematic review of existing research, followed by in-depth interviews with machine learning practitioners, subject matter specialists, and audit professionals for input on this topic. The system enhances its database through inclusion of new aspects such as control over information rights, transparency in decision-making processes, reliability and safety measures, moral considerations, and ongoing surveillance capabilities. Additionally, we create an effective auditing form containing over forty specific inquiries tailored to meet those requirements. Next, our focus is on integrating an expanded audit process within the machine learning development cycle of enterprises, including its advantages, obstacles, and ongoing research efforts

Ähnliche Arbeiten