Unlocking Potential of ‘Federated’ Approach in Machine Learning: Insights from Duke Professor

Jiaming Xu, an associate professor at Duke’s Fuqua School of Business, has highlighted the potential of a new approach to Artificial Intelligence (AI) systems training called federated learning. Xu said the approach works by training a central model using inputs received from decentralised sources, such as semi-autonomous cars, bank fraud detection systems and medical wearables. Xu said edge devices, such as smartphones, play a key role.
Xu said federated learning can work effectively because data from device to device is shared without it leaving the device, thereby minimising the risk of hacking and other privacy concerns. However, Xu admitted that communications between edge devices and servers create the potential for attacks such as eavesdropping. To combat this, Xu has developed strategies for optimising query complexity to enable secrecy while requiring the least amount of communication.