Liu, Xintong (University of Utah)
Faculty Advisor: Ji, Mingyue (College of Engineering, Department of Electrical and Computing Engineering)
One of the significant challenges of the machine learning faces today is how to deal with the privacy constraint of the user in a large-scale and distributed communication network. A myriad of data produced by billions of distributed devices need to be sent into the central cloud and to be managed, but what happens if the user does not want to send his/her data to the central cloud. It is reasonable that many users expect the data they send is being protected and maintain privacy. So, we are thinking about whether it is possible to create an individual machine learning in the application of Federated Learning so that the user's data will be protected from the privacy constraints. In this case, the raw data will not be known by anyone except the owner of these data. So, there would be all unknown input data pass through the private machine learning model, and the generated result, which is still hidden data will be sent back to the user. The main topic of the presentation is the designed codes which produces a private configuration with non-linear computation for the learning model and enable privacy constraints for the user's data.