WG5 - Federated learning

One of the unique features of our platform and collaboration is the possibility of federated learning. This is where data remains at local institutions but aggregate analysis or machine learning developed models can be iteratively learnt with equivalent results to if the data had all been stored at a single location. The AusCAT platform has initially been developed with horizontal learning, this is where data items for a single patient are all stored at a single location e.g. Mrs Smith’s data is all at institute A, Mr Nguyen’s data is all at institute B. Within this working group we will be reviewing and extending the learning algorithms and tools available for horizontal learning. We will also be working to extend this to include vertically partitioned data, which is where data items for a given patient are split across multiple institutions e.g. Mrs Smith has most data items at institute A but her medical data is stored at institute B. This will enable us to learn from federated data for multiple different scenarios and using multiple different algorithms and tools, to provide broad flexibility but still the required security and privacy.