Simplifying AI-Supported Development for Networking and Communication System

Date of Award


Degree Name

Ph.D. in Electrical and Computer Engineering


Department of Electrical and Computer Engineering


Feng Ye


Artificial Intelligence (AI)-based algorithms have demonstrated their robust capability to support networking and communication systems, such as network traffic classifier (NTC), intrusion detection systems, channel state information processing in massive multiple input multiple output (MIMO) wireless communication systems, etc. However, due to the relatively high-dimensional data input and limited computing resources, many of the existing AI implementations are too complicated for efficient processing in networking and communication systems. To address this issue, this dissertation explores a systematic approach that simplifies the AI-supported implementation for multiple networking and communication systems. The proposed approaches mainly evaluate the structure of AI implementations in different scenarios. In specific, for an AI-supported NTC development, an input feature contribution extraction scheme is developed to weigh each input feature based on both the significance and the uniqueness of the corresponding feature. The optimal set of input features is determined to minimize the complexity of targeting AI-based NTC while maintaining high performance in classification. Moreover, an autonomous update scheme is proposed to detect the changes in feature contribution and process updates. Evaluations of two fundamental AI-based classifiers demonstrated that the proposed scheme can significantly reduce the input features and accelerate NTC models by one to two magnitudes while maintaining high accuracy. The proposed autonomous update scheme can accurately detect a change in feature contributions and update the NTC models to sustain high accuracy. In addition, we further developed an adaptive pruning for MLP-based NTC to fit the different requirements of NTC due to network congestion. The results demonstrated that the lossless optimization and adaptive pruned network traffic classifier accelerate the baseline MLP-based NTC models by about 5 to 10 times based on the requirements. Besides NTC, this dissertation also developed a model simplification scheme targeting the deep learning based massive MIMO CSI feedback process in the next generation wireless communication systems. In this part, a dynamic channel sparsity scheme is proposed to optimize the optimized structure of the compact network. The lesser important channels and nodes are removed after the process for sustaining the reconstruction performance. Two popular deep learning based CSI feedback models are developed for evaluations. The results demonstrated that the proposed method can accelerate the baseline models by up to 3 times. Furthermore, this dissertation proposed a lossless optimization and simplification scheme. Using the MLP-based NTC as an example, the developed approach is to remove the nodes that contribute the least to the classification result in the hidden layer with the pruning method. Compared with the conventional pruning method, the proposed scheme can reduce the computational complexity while ensuring accuracy without retraining and fine-tuning. This dissertation has demonstrated that the current AI implementation in networking and communication systems can be simplified for high efficiency. The results have laid a solid foundation for future research in lightweight AI not only for the studied systems but also for a broader area that has limited computing resources and power supply.


Electrical Engineering, Deep learning, networking, communication system, data mining. structure pruning.

Rights Statement

Copyright © 2022, author.