Machine Learning Models Disclosure from Trusted Research Environments\n (TRE), Challenges and Opportunities

Abstract

Artificial intelligence (AI) applications in healthcare and medicine have\nincreased in recent years. To enable access to personal data, Trusted Research\nenvironments (TREs) provide safe and secure environments in which researchers\ncan access sensitive personal data and develop Artificial Intelligence (AI) and\nMachine Learning models. However currently few TREs support the use of\nautomated AI-based modelling using Machine Learning. Early attempts have been\nmade in the literature to present and introduce privacy preserving machine\nlearning from the design point of view [1]. However, there exists a gap in the\npractical decision-making guidance for TREs in handling models disclosure.\nSpecifically, the use of machine learning creates a need to disclose new types\nof outputs from TREs, such as trained machine learning models. Although TREs\nhave clear policies for the disclosure of statistical outputs, the extent to\nwhich trained models can leak personal training data once released is not well\nunderstood and guidelines do not exist within TREs for the safe disclosure of\nthese models.\n In this paper we introduce the challenge of disclosing trained machine\nlearning models from TREs. We first give an overview of machine learning models\nin general and describe some of their applications in healthcare and medicine.\nWe define the main vulnerabilities of trained machine learning models in\ngeneral. We also describe the main factors affecting the vulnerabilities of\ndisclosing machine learning models. This paper also provides insights and\nanalyses methods that could be introduced within TREs to mitigate the risk of\nprivacy breaches when disclosing trained models.\n

Ähnliche Arbeiten