👨🎓 About Me
LI Yuhui (Merrill Y. Li) is currently a first-year PhD student in the Department of Computing at The Hong Kong Polytechnic University, supervised by Prof. Jiannong Cao. Prior to that, he was graduated from College of Computer Science and Electronic Engineering at Hunan University (Supervisor: Prof. Wei Liang) in June 2024. He graduated from Shantou University with distinction (Rank 2/133, GPA 4.06/5.0) in July 2021 (Supervisor: Dr. Jianlong Xu).
I have a broad research interest in collaborative edge computing, network measurement, service computing, data mining, and deep learning. I have published several papers at the top computer conferences/journals such as IEEE ICME, IEEE INFOCOM, IEEE TC and IEEE TITS.
🔥 News
- 2024.03: One Paper Submitted - IEEE TSC (CCF-A).
- 2024.03: Major Revision Submitted - IEEE TDSC (CCF-A).
- 2023.05: One paper Accepted - IEEE SSE/SCC 2023 (CCF-C).
- 2022.12: One Paper Accepted - INFOCOM 2023 (CCF-A).
- 2022.09: One Paper Accpeted - ICPADS (CCF-C).
📝 Publications
📐 Network Measurement (WIP)
LEFT: Efficient Large Entries Retrieval in Network Monitoring
Wei Liang, Yuhui Li, Kun Xie, Naixue Xiong, Dafang Zhang, Kuan-Ching Li
Paper Summary
- My Roles : Idea, Code&Experiments, Writing
- Abstract : We observe that neural tensor completions (NTCs) usually bring better data recovery accuracy and recall performance under brute force searching. However, existing methods can not apply on NTCs and thus, one can not benefit from its better recall performance. This paper introduces LEFT, a differentiable tree-structured indexing model for efficient retrieval of large entries in network-wide monitoring data. LEFT leverages a max-heap tree and the beam search algorithm to determine the positions of top-K entries. Experimental results demonstrate the effectiveness and efficiency of LEFT, outperforming other retrieval baselines and achieving high scalability.
📐 Network Measurement
LightNestle: Quick and Accurate Neural Sequential Tensor Completion via Meta Learning
Yuhui Li, Wei Liang, Kun Xie, Dafang Zhang, Songyou Xie, Kuan-Ching Li
Pytorch Implementation PDF Download
Paper Summary
- My Roles : Idea, Experiments, Writing
- Abstract : For continous network measurement, the tensor is growing. When new data arrives, previous methods can not reach a perfect balance between accuracy and fast recovery. We are the first team to introduce meta-learning methods in tensor completion. We achieve significant improvement on both accuracy and training time.
EventMon: Real-Time Event-based Streaming Network Monitoring Data Recovery
Yuhui Li, Wei Liang, Kun Xie, Dafang Zhang, Kuan-Ching Li
Paper Summary
- My Roles : Idea, Experiments, Writing
- Abstract : Existing methods fail to satisfy the requirements of real-time recovery. We propose ResNMMF that achieves ultra-low latency data recovery in network measurement data streams as we process online data as streaming events. In addition, Stream2Batch is proposed to improve training and serving throughput. Extensive experiments show that ResNMMF achieves highly competitive accuracy under anytime inference settings.
LiteTC: Simplifying Neural Tensor Completion via Knowledge Distillation
Yuhui Li, Wei Liang, Kun Xie, Songyou Xie, Dafang Zhang, Kuan-Ching Li
Paper Summary
- My Roles : Idea, Experiments, Writing
- Abstract : The neural tensor completions adopted in network data recovery suffer from inefficient inference and difficulty in data retrieval. We propose LiteTC — a knowledge distillation framework that convert arbitrary advance model into a linear dot-product model, achieving both fast inference and retrieval friendly. Experiments demonstrate the conversion only suffer from reasonable recovery accuracy.
☁️ Service Computing
QoS Prediction and Adversarial Attack Protection for Distributed Services Under DLaaS
Wei Liang, Yuhui Li, Jianlong Xu, Zheng Qin, Dafang Zhang, Kuan-Ching Li
Paper Summary
- My Roles : Idea, Experiments, Writing
- Abstract : We show that the service recommendation system is suffering from prediction without confidence metric, cold start, and adversial vulnerability issues. We propose EEPrNN to address them and achieve confidence-aware, cold start tolerating and adversial robust service recommendation.
Inductive Subgraph Pattern Aware Model for Sparse QoS Prediction
Jianlong Xu, Zhiyu Xia, Yuhui Li, Zhidan Liu
Paper Summary
- My Roles : Idea, Experiments, Editing
- Abstract : We propose a novel subgraph-pattern GNN to inductively extract the rating signal from matrix-induced bipartite graph and enhance it by context-guided sampling. We demonstrate that our model largely outperforms previous methods when in high data sparsity. Our model achieve extrapolation on 25% unseen users/services with just 10% prediction accuracy loss.
QoSEraser: A Data Erasable Framework for Web Service QoS Prediction
Yuxiang Zeng, Yuhui Li, Zhiyu Xia, Zibo Du, Jialin Wang, Ruimin Lian, Jianlong Xu
Paper Summary
- My Roles : Idea, Editing
- Abstract : It is essential to provide users with the right to erasure regarding their own data, even if such data has been used to train a model, thanks to the GDPR. We propose a framework for QoS prediction, including collaborative signal preservation design and attentive embedding aggregation modules, to support fast retraining and mitigate the performance degradation.
NFMF:Neural Fusion Matrix Factorization for Personalized QoS Prediction in Service Selection
Jianlong Xu, Lijun Xiao, Yuhui Li, Mingwei Huang, Zicong Zhuang, Tien-Hsiung Weng, Wei Liang
Paper Summary
- My Roles : Idea, Experiments, Editing
- Abstract : We propose a context-aware neural matrix factorization for personalized service recommendation.
🚗 Urban Traffic
Spatial-temporal Aware Inductive Graph Neural Network for C-ITS Data Recovery
Wei Liang, Yuhui Li, Kun Xie, Dafang Zhang, Kuan-Ching Li, Alireza Souri, Keqin Li
Paper Summary
- My Roles : Idea, Experiments, Writing
- Abstract : The collected traffic monitoring data are usually incomplete due to unavoidable and unexpected incidents. Leveraging and modelling the neighborhood effect, we now support recovering missing data under all kinds of data missing patterns inductively without retraining.
ACSNet: Adaptive Cross-scale Network with Feature Maps Refusion for Vehicle Density Detection
Zuhao Ge, Yuhui Li, Cheng Liang, Youyi Song, Teng Zhou, Jing Qin
Paper Summary
- My Roles : Idea, Experiments, Writing
- Abstract : Using survellance camera to monitor traffic meets low-resolution, low-frame rate, and high occlusion problem. We introduce a novel ACSNet, leveraging adaptive cross-scale design and structural similarity regulation, to address the issues and achieve SOTA in two benchmarks.
💊 Bioinformatics
Predicting Drug-Target Interactions via Graph Differential Equations
Yuhui Li, Wei Liang, Li Peng, Dafang Zhang, Kuan-Ching Li
Paper Summary
- My Roles : Idea, Experiments, Writing
- Abstract : We jointly incorporate homogeneous graph convolution networks and heterogeneous graph neural networks to compensate the loss of information for each other. The experiments show that the proposed dual-stream encoder based design outperform the methods that only incorporate homogeneous or heterogeneous graph neural networks.
🎖 Honors and Awards
Scholarship
- 2023.11 1st-class Scholarship for Academic Excellence, HNU
- 2022.11 1st-class Scholarship for Academic Excellence, HNU
- 2021.11 1st-class Scholarship for Academic Excellence, HNU
- 2020.11 Outstanding Student Award, STU
- 2020.11 1st-class Scholarship for Academic Excellence, STU
- 2019.11 Mr. & Mrs. Yan Qingsen Scholarship
- 2018.11 2nd-class Scholarship for Academic Excellence, STU
- 2019.11 1st-class Scholarship for Academic Excellence, STU
- 2017.10 Scholarship for Outstanding New Students, STU
Awards
- 2020.10 2nd National Prize ACM “Lanqiao” Cup
- 2020.10 1st Prize ACM “Lanqiao” Cup in Province
- 2020.05 3rd Prize in the “Tedi” Data Mining Competition (Guangdong Province)
- 2019.12 3rd Prize in the Global Management Challenge
- 2019.11 2nd National Prize in the CUMCM
- 2019.09 1st Prize in the CUMCM (Guangdong Province)
- 2019.05 2nd Prize in the “Challenge Cup” Competition of Guangdong Province
💻 Internships/Visiting
- 2024.03 - 2024.06, Research Assistant, The Hong Kong Polytechnic University, Hung Hom, Kowloon Hong Kong.
📖 Educations
- 2021.09 - 2024.06, MEng, Hunan University, Changsha.
- 2017.09 - 2021.06, BEng, Computer Science, Shantou Univeristy, Shantou.
- 2014.09 - 2017.06, Foshan No.1 Middle School, Foshan.
💬 Invited Talks
- 2022.08 Deep Learning-based Missing Data Recovery