Security Issues and Solutions in Federate Learning Under IoT Critical Infrastructure

Digital world especially artificial intelligence and IoT have a vital and significant role in human life for exchanging the information and offering services through internet. These are recent and advanced technologies that has profound effects on various aspects of human life such as education, health care, socials and so on. IoT deals with various sensing objects and machines those interacts with each other for doing the activities. All these connected machines generates a huge amount of data and collects all data to one centralized server for further processing as well as managed by third party (cloud platform). Data has direct relationship with IoT devices however the size and number of IoT devices getting increase the size of collected data in central server (cloud) increase too so this has significant direct impact on performance, quality of services, computational power and time, the most important are data privacy and security issues respect to user data because centralized learning has classical algorithms to process IoT devices related data. More data owners were not interested to share their private and sensitive data with third organization cloud. Therefor to overcome these challenges and attract users’ attention the google founded system called federated learning is the affordable solution that enables on object-machine learning on which data owners and end users need not any more to share their sensitive data with others through centralized point cloud. Instead of centralized system they use distributed system and only need to share updated learning model between centralized server (cloud) and other end-user in the system. Federated learning (FL) boost the scope of decentralized approach because a huge number of data owners and end-users are interested to connect with each other to have a collaborated training without sending sensitive data to cloud. FL still vulnerable against various security threats and risks. This paper focus on two points; first we explore information about overall activity of IoT devices, structures and interactions with each other as well as information about cloud (centralized) learning system and distributed (FL) learning system. Second; the analysis of overall vulnerabilities being exploited by malicious attackers and security challenges to FL system and determine defensive techniques to enhance privacy and security in FL system to achieve security triad confidentiality, integrity and availability.