Optimising Scientific Collaboration with Federated Learning


External Member

Jacob Huang (Primary Supervisor)


Collaboration is a cornerstone of successful scientific research.  Individual researchers have specific skills, instruments in their lab and resources available to them.  This means a group of collaborators will each have unique data set, but also means that there are multiple ways they can contribute.  Is it better to share your samples, your expertise, or your data?  Is it possible to share nothing at all, and collaborate remotely by combining your results using federated learning?  In this project you will use a new federated learning package that includes federated data processing and fair aggregation, to compare different types of levels of collaboration between the digital twins of three hypothetical scientists (Alice, Bob and Charlie).   You will contribute to the code by revising it to train neural networks under different collaboration scenarios and compare the results. The code, digital twins and data sets will be provided.


Compare federated and non-federated learning models trained under different collaboration scenarios of three digital twins, to predict the best way to collaborate.


Python programming and experience in data science and machine learning is essential (such as COMP3720, COMP4660, COMP4670, COMP6670, COMP8420).  Familiarity with platforms such as Pytorch is desirable.


This is a single semester 12cp project.


machine learning, federated learning, neural networks

Updated:  10 August 2021/Responsible Officer:  Dean, CECS/Page Contact:  CECS Marketing