Federated learning is a machine learning approach for the training of artificial intelligence models distributed across multiple locations. In this context, the learning of features occurs at individual locations with the data available on-premise and thus data privacy is preserved. The collaborative learning experience in federated learning results in exceptional performance based on the comprehensive training data, which is more representative of the entire population participating in the learning process.
Recently, the federated learning approach has gained traction in the medical field. The combination of different sources of biomedical data have enabled more accurate clinical predictions, clinical outcomes and better model performance with strong privacy guarantees. We believe that federated learning has the potential to enrich the learning process of clinical data in different clinical applications. Our studies are focus on projects that accelerate applied research using federated learning in a medical environment. We are currently refining the optimizations at the core of the learning mechanisms and are evaluating our work with state-of-the-art methods for the segmentation of brain tumors.
Related publications:
- Reyes, J.*., Xiao,Y., & Kersten-Oertel, M. (2021). Data imputation and reconstruction of distributed Parkinson’s disease clinical assessments: A comparative evaluation of two aggregation algorithms [Accepted to MICCAI 2021 Workshop on Secure and Privacy-Preserving Machine Learning for Medical Imaging].
- Reyes, J.*., Di Jorio, L., Low-Kam, C., & Kersten-Oertel, M. (2021). Precision-Weighted Federated Learning. arXiv preprint arXiv:2107.09627.