Abstract:

The General Data Protection Regulation (GDPR) entered into force on 24th of May 2016 and applies since the 25th of May 2018. It imposes obligations on organizations independent on their involvement in commercial or research related activities as so long as they target or collect data related to people in the EU. While organizations are trying to adapt, the challenge lies in introducing GDPR related aspects in the academic curricula of even technical faculties such as computer science, electrical engineering, automation, etc.

In the above context, we are briefly introducing the PerHeart project which is developing a “Personalized ICT solution to reduce re-hospitalization rates in heart failure elderly patients suffering from comorbidities”. The project is proposing a novel approach for complying already at design level with the GDPR requirements. Federated Learning, a machine learning method which aims to train a model over a distributed data set, is implemented for all acquired health data. Federated Learning is taking into account restrictions on data confidentiality and the ability to operate over weak communications through networks. In PerHeart, we use an extension of the FedHealth method [2] that consists in clustering based on data distribution in order to perform distributed training. The evaluation of the proposed method was performed on two datasets i.e., HAPT and a sleep dataset. Both convergence and accuracy were improved compared to the traditional method.

In order to evaluate the knowledge of master level students on ethical constrains related to health data acquisition and processing, we have implemented a survey with 50 students from the Faculty of Artificial Intelligence, Computer Science Department, at the University Politehnica of Bucharest. The survey has provided insight on the educational level regarding the GDPR regulations and on the knowledge of the students regarding possible algorithms and methods which provide solutions for ethical compliance.

References:
[1] H. Brendan McMahan and Eider Moore and Daniel Ramage and Seth Hampson and Blaise Agüera y Arcas, Communication-Efficient Learning of Deep Networks from Decentralized Data, 2016, arXiv:1602.05629, Proceedings of the 20 the International Conference on Artificial Intelligence and Statistics (AISTATS) 2017. JMLR: W&CP volume 54.
[2] Yiqiang Chen and Jindong Wang and Chaohui Yu and Wen Gao and Xin Qin, FedHealth: A Federated Transfer Learning Framework for Wearable Healthcare, 2019, arXiv:1907.09173.

Keywords:

eHealth, ethical aspects, data confidentiality, federated learning.