Federated learning (FL) is a machine learning technique that enables distributed devices to train a learning model collaboratively without sharing their local data. FL-based systems can achieve much stronger privacy preservation since the distributed devices deliver only local model parameters trained with local data to a centralized server. However, there exists a possibility that a centralized server or attackers infer/extract sensitive private information using the structure and parameters of local learning models. We propose employing homomorphic encryption (HE) scheme that can directly perform arithmetic operations on ciphertexts without decryption to protect the model parameters. Using the HE scheme, the proposed privacy-preserving federated learning (PPFL) algorithm enables the centralized server to aggregate encrypted local model parameters without decryption. Furthermore, the proposed algorithm allows each node to use a different HE private key in the same FL-based system using a distributed cryptosystem. The performance analysis and evaluation of the proposed PPFL algorithm are conducted in various cloud computing-based FL service scenarios.

With the technical advances in ubiquitous comput- ing and wireless networking, there has been an increasing need to capture the context information (such as the location) and to figure it into applications. In this paper, we establish the theoreti- cal base and develop a localization algorithm for building a zero- configuration and robust indoor localization and tracking system to support location-based network services and management. The localization algorithm takes as input the on-line measurements of received signal strengths (RSSs) between 802.11 APs and between a client and its neighboring APs, and estimates the location of the client. The on-line RSS measurements among 802.11 APs are used to capture (in real-time) the effects of RF multi-path fading, temperature and humidity variations, opening and closing of doors, furniture relocation, and human mobility on the RSS measurements, and to create, based on the truncated singular value decomposition (SVD) technique, a mapping between the RSS measure and the actual geographical distance. The proposed system requires zero-configuration because the on-line calibration of the effect of wireless physical characteristics on RSS measurement is automated and no on-site survey or initial training is required to bootstrap the system. It is also quite responsive to environmental dynamics, as the impacts of physical characteristics changes have been explicitly figured in the mapping between the RSS measures and the actual geographical distances. We have implemented the proposed system with inexpensive off-the-shelf Wi-Fi hardware and sensory functions of IEEE 802.11, and carried out a detailed empirical study in our division building. The empirical results show the proposed system is quite robust and gives accurate localization results (i.e., with the localization error within 3 meters).

A smart grid facilitates more effective energy management of an electrical grid system. Because both energy consumption and associated building operation costs are increasing rapidly around the world, the need for flexible and cost-effective management of the energy used by buildings in a smart grid environment is increasing. In this paper, we consider an energy management system for a smart energy building connected to an external grid (utility) as well as distributed energy resources including a renewable energy source, energy storage system, and vehicle-to-grid station. First, the energy management system is modeled using a Markov decision process that completely describes the state, action, transition probability, and rewards of the system. Subsequently, a reinforcement-learning-based energy management algorithm is proposed to reduce the operation energy costs of the target smart energy building under unknown future information. The results of numerical simulation based on the data measured in real environments show that the proposed energy management algorithm gradually reduces energy costs via learning processes compared to other random and non-learning-based algorithms.