Hao Xiang(向浩)

Hao Xiang(向浩)

A doctoral student

Publications

Edge Computing Empowered Tactile Internet for Human Digital Twin: Visions and Case Study

Published in magazine 1, 2023

Tactile Internet (TI), with the capability of providing multisensory haptic human-machine interactions, is envisioned as a key enabling technology for an emerging application, called human digital twin (HDT). HDT is expected to revolutionize the human lifestyle and prompts the development of Metaverse. However, the realization of HDT poses stringent demands on pervasive connectivity, real-time feedback, high-fidelity modeling and ultra-high reliability (between physical and virtual spaces), which can hardly be met by TI only. In this article, we thus shed light on the design of edge computing empowered TI (ECoTI) for HDT. Aiming at offering strong interactions and extremely immersive quality of experience, we introduce the system architecture of ECoTI for HDT, and analyze its major design requirements and challenges. Moreover, we present core guidelines and detailed steps for system implementations. In addition, we conduct an experimental study based on our recently built testbed, which shows a particular use case of ECoTI for HDT in physical therapy, and the results indicate that the proposed framework, i.e., ECoTI, can significantly improve the effectiveness of the system. Finally, we conclude this article with a brief discussion of open issues and future directions.

Recommended citation: *Hao Xiang*, Kun Wu, Jiayuan Chen, Changyan Yi, Jun Cai, Dusit Niyato, Xuemin, Shen. (2023). "Edge Computing Empowered Tactile Internet for Human Digital Twin: Visions and Case Study." Magazine 1. 1(1). http://xianghao.github.io/files/paper1.pdf](https://arxiv.org/pdf/2304.07454.pdf

Research on Recommender System Based on Curiosity Guided Identity Modification

Published in Conference 1, 2022

Faced with the problem of information overload of big data, multi-factor fusion is the key technology of recommendation systems. How to provide personalized products for users accurately is the demand of recommendation system. Therefore, a new nearest neighbor algorithm is proposed to fuse the two kinds of identity and use curiosity as guidance to mining hidden information more efficiently, although the algorithm of curiosity modified identification degree swings in a small range, other evaluation indexes are improved. The improvement of the Receiver Operating Characteristic (ROC) curve shows that the robustness and improvement degree of the sub-algorithm is more significant.

Recommended citation: Xiang H, Dong Z, Gu P, et al. Research on Recommender System Based on Curiosity Guided Identity Modification[C]//2021 3rd International Conference on Artificial Intelligence and Advanced Manufacture. 2021: 333-340. https://dl.acm.org/doi/pdf/10.1145/3495018.3495075

PRSNet: A Masked Self-Supervised Learning Pedestrian Re-Identification Method

Published in Conference 1, 2022

In recent years, self-supervised learning has attracted widespread academic debate and addressed many of the key issues of computer vision. The present research focus is on how to construct a good agent task that allows for improved network learning of advanced semantic information on images so that model reasoning is accelerated during pre-training of the current task. In order to solve the problem that existing feature extraction networks are pre-trained on the ImageNet dataset and cannot extract the fine-grained information in pedestrian images well, and the existing pre-task of contrast self-supervised learning may destroy the original properties of pedestrian images, this paper designs a pre-task of mask reconstruction to obtain a pre-training model with strong robustness and uses it for the pedestrian re-identification task. The training optimization of the network is performed by improving the triplet loss based on the centroid, and the mask image is added as an additional sample to the loss calculation, so that the network can better cope with the pedestrian matching in practical applications after the training is completed. This method achieves about 5% higher mAP on Marker1501 and CUHK03 data than existing self-supervised learning pedestrian re-identification methods, and about 1% higher for Rank1, and ablation experiments are conducted to demonstrate the feasibility of this method. Our model code is located at this https URL.

Recommended citation: Xiao, Zhijie et al. “PRSNet: A Masked Self-Supervised Learning Pedestrian Re-Identification Method.” ArXiv abs/2303.06330 (2023): n. pag. https://arxiv.org/pdf/2303.06330.pdf

Research on State Modeling of Multiple Parsers Based on Attention Mechanism

Published in Conference 1, 2022

Transition-based dependency parsing is a fast and effective approach for dependency parsing. When the sentence is input to the transition-based dependency parser, the parser predicts a series of parsing actions from left to right. Stack information, which plays an essential role in the parsing process. In this paper, we propose a dependency parser that is based on bidirectional-LSTMs and an approach that uses the attention mechanism for learning representations of parser states. This model simulates the global state of the parser by capturing more relevant information, we dig into the stack information, buffer information, the word information that has been parsed out of the stack, the complete historical information of the actions taken by the parser, and semantic information during the parsing process. The words popped from the stack at each time step and the predicted action are then applied to the parsing task at the next time step, which is helpful to the parsing performance of the parser.

Recommended citation: B. Li, H. Xiang, C. Xiangxiu and N. Qun, "Research on State Modeling of Multiple Parsers Based on Attention Mechanism," 2021 2nd International Conference on Intelligent Computing and Human-Computer Interaction (ICHCI), Shenyang, China, 2021, pp. 198-203. https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=9708445

Yak Object Detection Based on Data Augmentation of Style Transfer Method

Published in Conference 1, 2022

To address the issue of complex yak data collection in the plateau area, as well as a lack of data, which leads to the limitation of the object detection model, a data enhancement method based on style transfer is used to increase the number of Tibetan plateau yak samples and improve object detection accuracy. In this study, we examine the results of several generative adversarial networks using the cycle generative adversarial network technique of alternate insertion residual network. Extend the original 450 yak data set by two times, manually generate four different data sets, compare the accuracy of different data sets using the YOLOv3 [1] object detection model, and verify that the alternating insertion residual network recurrent generation counter network improves the data effect. The test results suggest that this strategy may significantly enhance item detection accuracy in small samples.

Recommended citation: P. Gu, Z. Dong, Y. Xiao and H. Xiang, "Yak Object Detection Based on Data Augmentation of Style Transfer Method," 2021 International Conference on Information Technology and Biomedical Engineering (ICITBE), Nanchang, China, 2021, pp. 243-247, doi: 10.1109/ICITBE54178.2021.00060. https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=9694361

Big data cloud platform server load balancing algorithm based on improved chaotic partition algorithm

Published in Journal 1, 2021

Due to the heterogeneity of big data cloud platform and the uneven distribution of data between cloud servers, when cloud server cluster processes a large number of tasks, the node load is often uneven. To solve this problem, a load balancing algorithm based on improved chaotic partition algorithm for big data cloud platform server is proposed. According to the statistics of the average resource consumption of various services provided by the cluster, combined with the running time and resource occupation of tasks on the server, the total remaining task load of the server at a certain time point is predicted, so as to obtain the actual task load status of the node, and correct the task load in time. Experiments show that the load balancing algorithm of big data cloud platform server based on improved chaotic partition algorithm can effectively balance the load of multi task heterogeneous cloud servers, and has high feasibility, Then, based on this improved load balancing algorithm, we can also extend it to the application of multi-objective algorithm, such as robot path planning, target allocation scheduling algorithm and so on. There will be similar and reasonable performance compared with the same period last year.

Recommended citation: Xiang H , Zhang T , Li Z .Big data cloud platform server load balancing algorithm based on improved chaotic partition algorithm[J].Journal of Physics: Conference Series, 2021, 1982(1):012116-.DOI:10.1088/1742-6596/1982/1/012116. https://iopscience.iop.org/article/10.1088/1742-6596/1982/1/012116/pdf