OnCo
technologiesTechnologyEmerging

Federated learning and privacy-preserving AI

Training AI models across many hospitals without moving patient data, so the model learns from everyone while the data stay put.

Federated learning (NVIDIA FLARE, Owkin's Substra, Rhino Health, Intel OpenFL) trains a shared model on data held locally at each institution; used for pathology and radiology models (Owkin-led projects, the EXAM COVID model, Flywheel), and for pharma consortia (MELLODDY). Complementary tools include differential privacy, synthetic data (MDClone, Syntegra), and trusted execution environments. Governance and validation on heterogeneous data are the practical challenges.

Generic schematic · not to scale · placeholder for the ai computation front
Flagged finding · Neural network

How it works

Model updates, not data, are exchanged and aggregated centrally; privacy techniques limit what updates can reveal.

Strengths
  • Access to diverse, multi-site data
  • Regulatory and ethical acceptability
Limitations
  • Engineering and governance overhead
  • Non-identical data distributions
  • Still requires site IT capacity

Latest papers

top
Latest papers · live from Europe PMC
Open in Europe PMC

Query for this technology: (TITLE:"Federated learning and privacy-preserving AI" OR ABSTRACT:"Federated learning and privacy-preserving AI") AND (cancer OR tumor OR tumour OR oncology OR carcinoma OR lymphoma OR leukemia OR leukaemia OR myeloma OR sarcoma OR melanoma OR glioma). Results are unfiltered search hits about Federated learning and privacy-preserving AI, not a curated reading list.

Connected

7top