Federated deep learning for edge computing (part I)

• Main Author: See, Ian Soong En
• Other Authors: Tan Rui
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2020
• Subjects: Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence; Engineering::Computer science and engineering::Computer systems organization::Computer system implementation
• Online Access: https://hdl.handle.net/10356/138113
• Institution: Nanyang Technological University

With the increase in various usages of AI, comes new forms of training and deployment. One such advancement is coined as ‘federated learning’. Federated learning is an environment which consists of a central node that is connected through a network setting to multiple edge nodes to enable asynchronous model training. A main advantage of the federated learning framework is that data privacy of edge nodes is preserved throughout training. However, there are certain problems that are specific to Federated learning. One of these problems is network cost; of which this combined project hopes to solve by creating an efficent scheduling algorithm. This algorithm is being presented in part II of this project. The main purpose of this report is to detail the step by step process in the creation of a virtual environment that be used as a platform that supports testing and developing algorithms within federated learning. The virtual environment was created using the Mininet virtual image in conjunction with the Pysyft library. The output of part I of this project creates a virtual image (OVA format) that can be launched on a hypervisor such as oracle virtual box.