UPM Researchers Unveil Federated AI System for Hospital Data Privacy

What Happened

Researchers at the Universidad Politécnica de Madrid (UPM) have presented FedSDS, a federated learning strategy designed to allow hospitals to conduct collaborative analysis of patient data without transferring that data outside individual institutions. The system was announced as a response to longstanding tension between the clinical value of large-scale health data analysis and the legal and ethical constraints that restrict how patient records can be shared across organisations.

What FedSDS Does

Federated learning is a machine learning approach in which a model is trained across multiple decentralised devices or servers, each holding local data, without that data leaving its original location. Rather than pooling patient records into a central database, FedSDS allows each participating hospital to train on its own data and share only model updates, such as adjusted parameters or gradients, with a coordinating system.

UPM describes FedSDS as a new strategy within this framework, though full technical specifications of the system, including the architecture, the datasets used in validation, and the performance benchmarks, were not detailed in the available wire reports. The institution has not yet published peer-reviewed findings through the sources cited.

Background

Federated learning in healthcare has been an active area of research for several years. The approach was popularised in part by Google's early work on the technique and has since been applied to medical imaging, electronic health record analysis, and genomics, among other domains.

The primary regulatory driver in Europe is the General Data Protection Regulation, which places strict limits on the processing and transfer of sensitive personal data, including health information. Hospitals operating across different jurisdictions have faced particular difficulty sharing data for research purposes, even when doing so could improve diagnostic models or accelerate drug discovery.

UPM is a major public technical university based in Madrid, Spain, with research activity across engineering, computing, and applied sciences. It has not been identified in recent wire reports as a previous contributor to federated learning systems at clinical scale.

What It Means in Practice

If validated at clinical scale, a system like FedSDS could allow hospital networks to train shared AI diagnostic models using patient populations from multiple sites, improving the statistical robustness of those models without requiring any institution to expose individual records.

This matters particularly for rare disease research, where no single hospital accumulates sufficient patient volume to train a reliable model independently. It also has implications for cross-border research collaborations within the European Union, where data sovereignty requirements have historically slowed joint studies.

The system is presented as addressing both the technical and regulatory dimensions of the problem. Federated approaches do not eliminate all privacy risk, as research has shown that model updates can in some cases be reverse-engineered to infer properties of the underlying training data, but they substantially reduce the attack surface compared with centralised data aggregation.

What Remains Unknown

The wire reports do not specify whether FedSDS has been tested in live hospital environments or remains at the prototype and simulation stage. No hospital partners have been named. The timeline for any clinical pilot or external validation has not been disclosed. It is also not clear whether UPM is seeking commercial partners or intends to release the system as open-source research infrastructure.

No regulatory submissions have been reported in connection with FedSDS, and the system does not appear to have been submitted for review by European health technology assessment bodies.

What Comes Next

UPM has not announced a scheduled publication date for peer-reviewed findings, but the presentation of FedSDS suggests the researchers intend to pursue external validation and, potentially, pilot deployment within hospital networks in Spain or the broader European Union.