2° Seminar
Privacy and Federated Learning with Decision Trees
Speaker: Saloni Kwatra
Wednesday, October 24, 2023, at 2:00 PM CET
Department of Computer Science of the University of Pisa - Sala Seminari Ovest
Register in advance for this webinar on Microsft TEAMS
Abstract
In traditional machine learning approaches, the data is collected and stored in a single location for model training. Federated Learning (FL) addresses this issue by allowing the training of a shared model across multiple distributed devices or organizations without the need for centralized data collection.
In FL, the data remains with the local devices or organizations, and only the model updates are shared with the central server. The central server aggregates the model updates received from different devices and sends the aggregated model updates back to the distributed devices. The process is continued until the model reaches a point of convergence or until the maximum number of iterations has been achieved. For e.g., the Federated Averaging (FedAvg) algorithm was proposed to build deep learning models in an FL framework.
Our work focuses on building FL frameworks with Decision Trees (DTs), as DTs are easy-to-interpret machine learning models, consisting of nodes with a split attribute and a split value organized as a tree. Although the data is not shared among the distributed devices and the aggregation server, sharing model parameters leads to substantial privacy breaches in the FL framework. Therefore developing FL frameworks, which also preserve the privacy of individuals, is necessary.
--------------------------------------
The "Colorful Seminars: Enhancing Diversity and Inclusion" is an initiative of UNIPI to increase the representativeness of diverse groups and bridge the equity gap between underrepresented and marginalized communities.
International experts belonging to different communities will be invited to hold seminars on the topics covered by the SoBigData.it project to promote more diverse and inclusive participation in computer and data science events.
The seminars will be hosted by the Department of Computer Science (also with streaming) or will be held exclusively online.
