This paper proposes a novel approach for multi-party collaborative data analysis problems, where analysis accuracy and divergence are required, as well as both privacy of shared data and explainability of results. The proposed approach aims at trading-off data privacy, decision explainability, and data utility by analytically relating these three measures, evaluating how they impact each other, and proposing a methodology to find the best possibletrade-offamong them. In particular, given a set of requirements from the participants for a collaborative analysis problem, we propose a method to properly tune the parameters of privacy-preserving mechanisms and explainability techniques to be adopted by all participants, obtaining the besttrade-off. The paper is focused on deep learning-based image data analysis problems, though the approach can be generalized to other data types. The$(\epsilon , \delta )$-Differential Privacyand theAutoencodersprivacy-preserving techniques have been adopted to preserve data privacy, while theSmoothGradmechanism has been used to provide decision explainability. The proposed methodology has been validated with a set of experiments on three multi-class deep learning classifiers and three well-known image datasets, MNIST, FER, and CIFAR-10.

In this paper, we present further contributions to the field of privacy-preserving and explainable data analysis applied to tabular datasets. Our approach defines a comprehensive optimization criterion that balances the key aspects of data privacy, model explainability, and data utility. By carefully regulating the privacy parameter and exploring various configurations, our methodology identifies the optimal trade-off that maximizes privacy gain and explainability similarity while minimizing any adverse impact on data utility. To validate our approach, we conducted experiments using five classifiers on a binary classification problem using the well-known Adult dataset, which contains sensitive attributes. We employed (ϵ, δ)-differential privacy with generative adversarial networks as a privacy mechanism and incorporated various model explanation methods. The results showcase the capabilities of our approach in achieving the dual objectives of preserving data privacy and generating model explanations.

This paper proposes a novel approach for privacy preserving face recognition aimed to formally define a trade-off optimization criterion between data privacy and algorithm accuracy. In our methodology, real world face images are anonymized with Gaussian blurring for privacy preservation. The anonymized images are processed for face detection, face alignment, face representation, and face verification. The proposed methodology has been validated with a set of experiments on a well known dataset and three face recognition classifiers. The results demonstrate the effectiveness of our approach to correctly verify face images with different levels of privacy and results accuracy, and to maximize privacy with the least negative impact on face detection and face verification accuracy.

In this paper, we demonstrate an application of privacy-preserving face recognition combined with an Attribute-Based Access Control framework to regulate access from subjects to critical resources while preserving the subject's privacy. The demonstrator exploits a mechanism that dynamically computes the best trade-off between ensured privacy and data utility, based on image acquisition conditions, and a decision engine based on XACML policies to express complex and dynamic conditions. The demonstrator can handle the dynamic association of new identities, as well as modification of access conditions. Attendees of the demo session can interact with the demo in a variety of ways, including modifying the camera input, but also through the customization of rules as well as the privacy parameter.