< BK21+ BEST Seminar Series Announcement> 

Time and Date : 11:00 ~ 12:00 Saturday 06/22/2019

Place : D403, Engineering Building #4

Title : Communication-Efficient Federated Machine Learning Frameworks
Abstract:
Recent advances in mobile computing power and machine learning (ML) have spurred on to the emergence of intelligent devices at the wireless network edge, ranging from phones and cameras to cars, drones, and robotic assemblers in smart factories. These devices generate private data by observing the surrounding environments, and run on-device ML models such as neural networks, thereby locally carrying out mission-critical decisions promptly. Since each single device observes a tiny fraction of the global environmental data, the on-device ML models should be trained collectively across devices over wireless links. This calls for a novel distributed ML framework that is suited for wireless connectivity while preserving local data privacy, termed federated machine learning (FML). In this talk, promising communication-efficient FML frameworks will be outlined, wherein on-device ML models are trained by their exchanging: model parameters, model outputs, and/or local data with privacy guarantees. Their effectiveness will be discussed in supervised, unsupervised, and reinforcement learning applications, under different channel dynamics and network architectures, as well as data distributions and privacy guarantee requirements.

Presenter: Jihong Park & Post Doctoral Researcher / University of Oulu, Finland

Host: Prof. Kim, Seonglyun, Yonsei EEE