Abstract<\/p>\n
Federated Learning (FL) is a recent direction in Machine Learning. Whereas standard learning approaches require centralizing all the training data on one machine, FL enables training with decentralized data privately stored across many machines. This is important because in practice data may be too big to send to the central server or too sensitive to share. However, for FL to fulfill its potential, three research questions arise: how to (1) maximize learning accuracy given the statistical heterogeneity in the local training data, (2) incentivize contributions of good local training data, and (3) minimize FL\u2019s dependence on any central authority that could be prone to fraud and performance bottleneck. This project is an ambitious effort to address these questions by leveraging blockchain and edge computing. The results will benefit many learning applications; for example, those involving self-driving cars, medical records, AI cameras, and mobile user behaviors.<\/p>\n
Authors: Duc Tran<\/p>\n","protected":false},"excerpt":{"rendered":"
Abstract Federated Learning (FL) is a recent direction in Machine Learning. Whereas standard learning approaches require centralizing all the training data on one machine, FL enables training with decentralized data privately stored across many machines. This is important because in practice data may be too big to send to the central server or too sensitive […]<\/p>\n","protected":false},"author":2,"featured_media":425,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":[],"categories":[3],"tags":[37],"acf":[],"yoast_head":"\n