Federated Learning and Privacy

Federated Learning is a distributed learning paradigm which arises from both, the increasing awareness of data privacy and the growing number of smart devices connected to the internet. To sum up, Federated Learning consists of a set of clients which agree to train a learning model in a collaborative way under the orchestration of a central server. Although the learning process is orchestrated by a central server, the training data of each client is never shared so it increases their privacy. This is the main difference with the classic distributed learning setup.

Even though, client’s training data is not shared, they share their trained models and there are several ways to extract additional information from such models. To ensure a higher degree of privacy many Differential Privacy techniques can be employed.

There are many open problems and challenges in this field, such as: adversarial attacks and defense mechanisms, personalization, adapting classic machine learning models to this federated setup and so on.

Contact: María Victoria Luzón García

Related Researchers:

Letra:

  Name Email Area Cat.
Herrera Triguero, Francisco herrera@decsai.ugrhDP7shEY7UM.es DaSCI Technology Applications Area, Data Science and Big Data Area, Computational Intelligence Area PhD
Luzón García, María Victoria luzon@ugCmlk63ibwjUOr.es DaSCI Technology Applications Area PhD
Luzón García, María Victoria luzon@uVifDNngr.es DaSCI Technology Applications Area PhD
Martínez Cámara, Eugenio emcamara@decsai.ugxU.ua.er.es DaSCI Technology Applications Area PhD - Juan de la Cierva
Peregrin Rubio, Antonio peregrin@dg.pqi_0ti.uhu.es Computational Intelligence Area PhD
Peregrin Rubio, Antonio peregrin@dti.uhu.0wRC02GSubres Computational Intelligence Area PhD
Scroll Up