Federated Learning for Privacy-Preserving AI Model Training

Abstract

Artificial Intelligence (AI) has been used variably in healthcare diagnostics thereby demonstrating promising results especially in the way diagnosis is done. However, one of the strategies that appear to hinder its mass implementation is the lack of traceability of the process leading to the generation of the result through the AI models, thus creating a trust deficit between the medical practitioners and AI systems. In this paper, I discuss Explainable Artificial Intelligence (XAI) with an aim disentangling the diagnostic domain of the healthcare system to improve the performance of AI models while making them more explainable. We overview existing methods in the field of XAI, the use of these techniques in the healthcare system and the need to develop trust between AI systems and healthcare workers. In addition, we outline recommendations for how to close this trust divide, which includes both understandable instructions and the benefits of the implementation of XAI in rendering health care services.

Ähnliche Arbeiten