Machine Learning Projects

ML-based Matrix Optimization in Massive MIMO

In the downlink of massive MIMO, the transmitter uses precoding technology to reduce interference and improve spectrum efficiency. A complex-valued gradient neural network (CVGNN) is proposed to solve the Moore-Penrose inverse of the complex matrix in massive MIMO precoding algorithms. Theoretical linear convergence and numerical results are provided to corroborate the application of CVGNN in wireless communication senarios.

• Undergraduate Thesis, Southeast University.

Fog Computing in Internet of Vehicles

As the technology of network, wireless transmission and big data computing develops rapidly, the society has entered the Information Era. Internet of Vehicles, the important part of the Internet of Things, is the inevitable trend of urban traffic in the future. Fog computing emerges as a new technology of distributed computing, which is suitable for applications in the Internet of Vehicles. Internet of Vehicles based on fog computing can solve the problems of traffic congestion, transportation efficiency and security. In this paper, we review the recent work in this topic and analyze the architecture and scenario of fog computing in the Internet of Vehicles. Three aspects of applications are proposed in this paper: VANET, big data processing and security. At the same time, we discuss the technical details and challenges in these applications. Fog computing has excellent processing power in real-time big data and a high mobility environment, which reduces the latency of data processing, makes the deployment between vehicles and roadside units more reasonable and enhances the security of information interchanged in the Internet of Vehicles. Fog computing devices will become an indispensable part of the future construction of Internet of Vehicles.

• “A Survey on Fog Computing Applications in Internet of Vehicles”, International Conference on Computing and Data Science, 2021.

Online Review Classifier

Customer reviews on e-commerce platforms contain valuable information, while sifting through them manually tends to dismay people because of the huge amount of data. This study implemented a machine learning-based algorithm to classify customer reviews. Our classifier extracts Chinese word segmentation and text frequency for feature extraction and scoring, and implements the classification with methods of Naive Bayesian and Support Vector Machines. Experimental results on the Taobao product review sentiment datasets show that our model based on two machine learning algorithms, though results in different performances, can provide suggestions on the selection of the identification classifier using a trade-off strategy and helps obtain fast and accurate classification on reviews of different categories.

• “Classification of Customer Reviews on E-commerce Platforms Based on Naive Bayesian Algorithm and Support Vector Machine”