Search

In federated learning (FL), model parameters of deep learning are communicated between clients and the central server. To better train deep learning models, the spectrum resource and transmission security need to be guaranteed. Toward this end, we propose a secrecy transmission protocol based on energy harvesting and jammer selection for FL, in which the secondary transmitters can harvest energy from the primary source. Specifically, a secondary transmitter ST i is first selected, which can offer the best transmission performance for the secondary users to access the primary frequency spectrum. Then, another secondary transmitter ST n , which has the best channel for eavesdropping, is also chosen as a friendly jammer to provide secrecy service. Furthermore, we use outage probability (OP) and intercept probability (IP) as metrics to evaluate performance. Meanwhile, we also derive closed-form expressions of OP and IP of primary users and OP of secondary users for the proposed protocol, respectively. We also conduct a theoretical analysis of the optimal secondary transmission selection (OSTS) protocol. Finally, the performance of the proposed protocol is validated through numerical experiments. The results show that the secrecy performance of the proposed protocol is better than the OSTS and OCJS, respectively. [ABSTRACT FROM AUTHOR]

Human Activity Recognition (HAR) employing inertial motion data has gained considerable momentum in recent years, both in research and industrial applications. From the abstract perspective, this has been driven by an acceleration in the building of intelligent and smart environments and systems that cover all aspects of human life including healthcare, sports, manufacturing, commerce, etc. Such environments and systems necessitate and subsume activity recognition, aimed at recognizing the actions, characteristics, and goals of one or more individuals from a temporal series of observations streamed from one or more sensors. Due to the reliance of conventional Machine Learning (ML) techniques on handcrafted features in the extraction process, current research suggests that deep-learning approaches are more applicable to automated feature extraction from raw sensor data. In this work, the generic HAR framework for smartphone sensor data is proposed, based on Long Short-Term Memory (LSTM) networks for time-series domains. Four baseline LSTM networks are comparatively studied to analyze the impact of using different kinds of smartphone sensor data. In addition, a hybrid LSTM network called 4-layer CNN-LSTM is proposed to improve recognition performance. The HAR method is evaluated on a public smartphone-based dataset of UCI-HAR through various combinations of sample generation processes (OW and NOW) and validation protocols (10-fold and LOSO cross validation). Moreover, Bayesian optimization techniques are used in this study since they are advantageous for tuning the hyperparameters of each LSTM network. The experimental results indicate that the proposed 4-layer CNN-LSTM network performs well in activity recognition, enhancing the average accuracy by up to 2.24% compared to prior state-of-the-art approaches. [ABSTRACT FROM AUTHOR]

As the Internet of Things (IoT) is predicted to deal with different problems based on big data, its applications have become increasingly dependent on visual data and deep learning technology, and it is a big challenge to find a suitable method for IoT systems to analyze image data. Traditional deep learning methods have never explicitly taken the color differences of data into account, but from the experience of human vision, colors play differently significant roles in recognizing things. This paper proposes a weight initialization method for deep learning in image recognition problems based on RGB influence proportion, aiming to improve the training process of the learning algorithms. In this paper, we try to extract the RGB proportion and utilize it in the weight initialization process. We conduct several experiments on different datasets to evaluate the effectiveness of our proposal, and it is proven to be effective on small datasets. In addition, as for the access to the RGB influence proportion, we also provide an expedient approach to get the early proportion for the following usage. We assume that the proposed method can be used for IoT sensors to securely analyze complex data in the future. [ABSTRACT FROM AUTHOR]

Mobile cellular communications are experiencing an exponential growth in traffic load on Long Term Evolution (LTE) eNode B (eNB) components. Such load can be significantly contained by directly sharing content among nearby users through device-to-device (D2D) communications, so that repeated downloads of the same data can be avoided as much as possible. Accordingly, for the purpose of improving the efficiency of content sharing and decreasing the load on the eNB, it is important to maximize the number of simultaneous D2D transmissions. Specially, maximizing the number of D2D links can not only improve spectrum and energy efficiency but can also reduce transmission delay. However, enabling maximum D2D links in a cellular network poses two major challenges. First, the interference between the D2D and cellular communications could critically affect their performance. Second, the minimum quality of service (QoS) requirement of cellular and D2D communication must be guaranteed. Therefore, a selection of active links is critical to gain the maximum number of D2D links. This can be formulated as a classical integer linear programming problem (link scheduling) that is known to be NP-hard. This paper proposes to obtain a set of network features via deep learning for solving this challenging problem. The idea is to optimize the D2D link schedule problem with a deep neural network (DNN). This makes a significant time reduction for delay-sensitive operations, since the computational overhead is mainly spent in the training process of the model. The simulation performed on a randomly generated link schedule problem showed that our algorithm is capable of finding satisfactory D2D link scheduling solutions by reducing computation time up to 90% without significantly affecting their accuracy. [ABSTRACT FROM AUTHOR]