"Will Federated Learning Revolutionize AI Training?"
Jiaming Xu, an associate professor at Duke University's Fuqua School of Business, and his coauthors explored how to keep data safe and private when using a new decentralized, collaborative way of training Artificial Intelligence (AI) models. Xu says federated learning is a new approach that can revolutionize AI systems training. He explains that traditional Machine Learning (ML) requires all data to be centralized in a single data center, while federated learning, also known as collaborative learning, trains the algorithm of a central model using data from decentralized sources. Edge devices are crucial in federated learning. These data-collecting tools, including smartphones, climate sensors, semi-autonomous vehicles, satellites, bank fraud detection systems, and medical wearables, could remotely share their data to train the central learning model in a repeatable cycle. According to Xu, federated learning is appealing to scientists, doctors, and companies because the data itself never leaves the edge devices. Due to privacy laws such as the Health Insurance Portability and Accountability Act (HIPAA), and the threat of hacking, federated learning is attractive to various industries. However, according to Xu, federated learning is not a perfect solution since the possibility of hacking still exists when the server and edge devices communicate. It is still possible for an eavesdropper to infer the private data based on the sent parameters. Therefore, Xu has developed querying strategies and analytical techniques that can be incorporated into theft-proof frameworks for federated learning in order to help find privacy solutions. His findings are presented in two papers titled "Learner-Private Convex Optimization" and "Optimal Query Complexity for Private Sequential Learning Against Eavesdropping." This article continues to discuss the concept of federated learning and the research on privacy solutions for this approach.