MPC-Learning consists of a distributed computation model designed to preserve data-privacy and -confidentiality, enabling machine learning models to be taken to where the data is instead of working with a single centralized dataset.

Anonymization is a way of cutting down the risks involved in mass data-collection and -processing. It hides the personal or sensitive data, enabling it to be broadcast without thereby violating data protection rights. Nonetheless, an anonymized database is prone to what are called re-identification attacks, meaning an attempt to trace ostensibly anonymous records in the records of another related database or data source to extract confidential information from it.

Mindful of this, GMV is taking part in the project “MPC-Learning: Aprendizaje automático seguro y protegido mediante compartición de secretos” (MPC-Learning. Secure and protected machine learning by secret sharing). Co-funded by the R&D department of GMV’s Secure e-Solutions sector and the Spanish Ministry of Economic Affairs and Digital Transformation (Ministerio de Asuntos Económicos y Transformación Digital), the project focuses on the development of mathematical and computational techniques capable of numerical calculation without the need for sharing data. In the words of Juan Miguel Auñón-García, Data Scientist of GMV’s Secure e-Solutions sector “GMV has a twofold role in this project. Firstly, it is inputting its wealth of experience in sectors like banking and healthcare, thus helping to build up use cases in order to solve our clients’ difficulties. Secondly, GMV is also bringing to the table its data science expertise, which underpins the secure performance of the computation. The synergies resulting from these two inputs enables us to take on these challenges with all due confidence of success”.

MPC-Learning springs from a premise: on many occasions several organizations might be interested in sharing information so that all of them can learn from each other, but there might be constraints to this exchange of information, such as applicable legislation or legitimate interest. Multi-Party Computation (MPC) pools individuals (parties) who wish to perform a computation without having to reveal their data but in due awareness that collaboration is essential in quest of a common good.

The project’s aim is to build a federated learning platform and a distributed computation model designed to preserve data privacy and confidentiality, in which participants can train up their Machine/Deep Learning models in a secure way, each participant learning from all the rest.