RMF: A Risk Measurement Framework for Machine Learning Models

TL;DR

This paper introduces RMF, a framework for assessing the security risks of machine learning models, especially in autonomous vehicles, using risk indicators based on ISO standards to evaluate potential damage and attacker effort.

Contribution

It presents a novel risk measurement framework tailored for ML security, incorporating multiple risk indicators and a case study application.

Findings

Framework effectively measures security risks in ML systems.

Risk indicators provide nuanced insights into attack efforts and damages.

Case study demonstrates practical applicability of RMF.

Abstract

Machine learning (ML) models are used in many safety- and security-critical applications nowadays. It is therefore important to measure the security of a system that uses ML as a component. This paper focuses on the field of ML, particularly the security of autonomous vehicles. For this purpose, a technical framework will be described, implemented, and evaluated in a case study. Based on ISO/IEC 27004:2016, risk indicators are utilized to measure and evaluate the extent of damage and the effort required by an attacker. It is not possible, however, to determine a single risk value that represents the attacker's effort. Therefore, four different values must be interpreted individually.