Cyrile Verdeyen (4th year MSci in Software Engineering Computer Science at Royal Holloway University of London) is working as a UROP student in the summer of 2019. Before coming to the UK to study he lived in Santiago de Chile, where he completed the IB.
He has called the project he is working on DECML, which stands for Distributed Edge Consensus Machine Learning. The idea stems from federated machine learning, where instead of data from edge devices is sent to a central server for machine learning, the machine learning is done on the devices themselves, and instead the model is sent to the central server. He is taking this concept one step further, and instead of having the edge nodes send their model back to the central server, the central server can send the nodes questions, which they can then answer with the model that node has. This way a node can share what it knows without sharing its own data or its machine learning model. This privacy-preserving model can be used in situations such as labeling what a patient may have at a hospital. Multiple hospitals agreed to work together using DECML, one of the hospitals is not sure is a patient has a certain disease or not, and so presents the question to the cluster orchestrator (CO). The CO than sends out the question for all the other Hospitals to classify it, and send back a log of all their combined answers to the CO, which takes a consensus of the answers, and sends it to the hospital which asked the question.