Page 59 - AI for Good Innovate for Impact
P. 59
AI for Good Innovate for Impact
Use Case 3: CAFEIN – Federated Learning Platform
Country: Switzerland 4.1-Healthcare
Organization: CERN
Contact Person(s):
Diogo Reis Santos, diogo.reis.santos@ cern .ch,
Luigi Serio, luigi.serio@ cern .ch
Alessandro Raimondo, a.raimondo@ cern .ch
1 Use Case Summary Table
Item Details
Category Healthcare
Problem Addressed In many industries, sensitive data is distributed across different insti-
tutions. Traditional centralized machine learning approaches require
data sharing that may compromise privacy, regulatory compliance, and
data sovereignty. There is a growing need for solutions that enable
collaborative analytics and AI model training without moving raw data.
Key Aspects of Solu- CERN's Federated Learning Infrastructure (CAFEIN) is a comprehen-
tion sive, fully hosted platform—not just a framework—that provides an
integrated software framework, robust software infrastructure, and
dedicated server-side hardware infrastructure.
The platform is hosted, operated, and managed by Conseil Européen
pour la Recherche Nucléaire (CERN), ensuring that it benefits from
CERN’s recognized network security, as well as its status as a non-profit
international organization with non-military application principles.
CAFEIN facilitates secure federated learning and analytics by enabling
decentralized model training while ensuring that sensitive data remains
local. Advanced privacy-preserving technologies, such as secure
aggregation and differential privacy, further protect data integrity and
confidentiality.
Technology Keywords Federated Learning, Federated Analytics, Federated Inference, Priva-
cy-preserving, Machine Learning, Artificial Intelligence
Data Availability Local data is maintained and secured by each participating organiza-
tion, ensuring data privacy and regulatory compliance.
Metadata (Type of The platform is data-type agnostic. It supports structured and unstruc-
Data) tured data, including numerical, text, and image data from various
domains (e.g., sensor data, operational logs, imaging).
23
