The proliferation of smart devices has caused data to be generated in an increasingly distributed manner. Due to high cost and privacy concerns, this data cannot necessarily be transferred to a centralized location for model training. Furthermore, computational resources and data distributions may vary substantially from location to location. We present a federated learning system that addresses these sources of heterogeneity and substantially reduces model training time in these distributed settings.
Joel Wolfrath is a third-year Ph.D. candidate in the Department of Computer Science & Engineering, advised by Dr. Abhishek Chandra. His interests include distributed systems, edge computing, and statistics.