Rethinking Certification for Trustworthy Machine Learning-Based Applications

TL;DR

This paper discusses the challenges of certifying non-deterministic ML applications, analyzes current shortcomings, and proposes a new certification scheme to improve trustworthiness in ML-based systems.

Contribution

It introduces the first certification scheme specifically designed for ML-based applications, addressing existing gaps in assurance methods.

Findings

Identifies key challenges in certifying ML applications

Highlights deficiencies of current certification schemes

Proposes a novel certification framework for ML systems

Abstract

Machine Learning (ML) is increasingly used to implement advanced applications with non-deterministic behavior, which operate on the cloud-edge continuum. The pervasive adoption of ML is urgently calling for assurance solutions assessing applications non-functional properties (e.g., fairness, robustness, privacy) with the aim to improve their trustworthiness. Certification has been clearly identified by policymakers, regulators, and industrial stakeholders as the preferred assurance technique to address this pressing need. Unfortunately, existing certification schemes are not immediately applicable to non-deterministic applications built on ML models. This article analyzes the challenges and deficiencies of current certification schemes, discusses open research issues, and proposes a first certification scheme for ML-based applications.