An Information Theoretic Approach to Privacy-Preserving Interpretable and Transferable Learning

In order to develop machine learning and deep models that take into account the guidelines principles of trustworthy AI, a novel information theoretic approach is introduced in this article. A unified privacy-preserving interpretable transferable considered for studying optimizing trade-offs between privacy, interpretability, transferability aspects AI. variational membership-mapping Bayesian model used analytical approximation defined measures privacy leakage, transferability. The consists approximating by maximizing lower-bound using optimization. demonstrated through numerous experiments on benchmark datasets real-world biomedical application concerned with detection mental stress individuals heart rate variability analysis.