A differential privacy based prototypical network for medical data learning

Abstract

The use of machine learning models in numerous computing domains is becoming increasingly widespread, and the number of machine learning models used in diverse fields is growing. The security issues of machine learning models are getting increasingly important as a result of continual in-depth study for medical data learning. This research provides a training model based on differential privacy that blends differential privacy with a prototypical network for medical data learning. We verify the performance of the proposed method on the open-source Omniglot dataset and a synthetic dataset from Dermnet website and Kaggle, and results demonstrate that our method achieves privacy protection during medical data learning without a big drop in classification accuracy.

Ähnliche Arbeiten