Federated Transfer Learning: Current Issues and New Perspectives
Abstract
Federated Learning is an advancement of Machine Learning where the model training is accomplished with a collection of decentralized local data maintaining the data privacy. Transfer Learning is the process of reusing the features learned with a pre-trained model to a different problem domain. We survey the literature on the cutting edge, Federated Transfer Learning which combines the characteristics of both Federated and Transfer Learning. We identify key technical issues in its application areas of Healthcare, Agriculture, City Management and Resource Allocation and discuss the challenges which includes System Heterogeneity, Expensive Communication, Model Poisoning Attacks and Model Aggregation. At last, we discuss the new opportunities including New Tools for Quantifying Heterogeneity, New Methods on Convergence, Bandwidth Efficiency and Collaborative Mobile Clustering Learning that are promising for future research.
Federated Transfer Learning: Current Issues and New Perspectives
Volstorff A
Federated Learning is an advancement of Machine Learning where the model training is accomplished with a collection of decentralized local data maintaining the data privacy. Transfer Learning is the process of reusing the features learned with a pre-trained model to a different problem domain. We survey the literature on the cutting edge, Federated Transfer Learning which combines the characteristics of both Federated and Transfer Learning. We identify key technical issues in its application areas of Healthcare, Agriculture, City Management and Resource Allocation and discuss the challenges which includes System Heterogeneity, Expensive Communication, Model Poisoning Attacks and Model Aggregation. At last, we discuss the new opportunities including New Tools for Quantifying Heterogeneity, New Methods on Convergence, Bandwidth Efficiency and Collaborative Mobile Clustering Learning that are promising for future research.