一、Overview：
Research Overview：
The Guangdong Provincial Engineering Technology Research Center for Information Access and Transmission Security is primarily dedicated to the fields of parallel acceleration and performance optimization for heterogeneous computing, software reliability assurance, and user behavior analysis. In the domains of building a domestic Artificial Intelligence ecosystem and enabling computing power, the center assists Huawei in the development and performance optimization of operator libraries for its Kunpeng + Ascend NPU computing platform. Benchmarking against the NVIDIA CUDA and AMD ROCM ecosystems, the center facilitates the migration of typical industry applications—such as ship face recognition and intelligent financial engines—to domestic platforms represented by Huawei. This work ensures that-key industry applications can be successfully transitioned to domestic AI computing platforms while achieving satisfactory performance metrics. In recognition of these contributions, the collaborative projects led by the center have repeatedly received the Excellent Cooperation Project award from Huawei. Furthermore, the center was consecutively named MVP by Huawei's Computing Product Line for the years 2023 and 2024.
Scientific Achievements：
The Engineering Center has recently published over 10 research papers, filed 21 patent applications (14 granted), and applied for 8 software copyrights.
Research Institutions And Platforms：
Guangdong Provincial Engineering Technology Research Center for Information Access and Transmission Security.
Designation: Guangdong Provincial Engineering Center
Technology Transfer：
Leveraging mainstream domestic computing platforms such as Ascend and Sugon, the center conducts research and development in several key areas. This includes the design and open-sourcing of foundational operator template libraries and algorithms for long-text sequences in Large Language Models (LLMs). Notably, the optimized Cutlass template library has achieved a 1.4x performance speedup compared to its counterpart on the NVIDIA A100. For the Shenzhen Supercomputing Center, the center has performed HPL Linpack DGemm optimization for narrow, small-shape matrices (384/1920) on the latest Kunpeng platform. It has also designed a shared memory algorithm for large-scale memory sharing libraries within inference environments. Furthermore, the center has developed a fusion algorithm for Ascend's FP8 and general-purpose computing, and its FP8 implementation successfully addresses the previous lack of FP8 support in the Ascend platform's inference environment. These efforts have enabled the successful migration of typical industry applications—such as ship face recognition, intelligent finance, and OPPO mobile phone rendering—to domestic artificial intelligence platforms, contributing significantly to ecosystem enablement.
二、Detail
Name：
Guangdong Provincial Engineering Technology Research Center for Information Access and Transmission Security.
Introduction：
The platform is managed and operated under the auspices of the School of Computer Science and Engineering at South China University of Technology. Its core mission is to address critical bottlenecks in the performance optimization of domestic computing power, focusing on three primary areas: parallel acceleration and performance optimization for heterogeneous computing, software reliability assurance, and user behavior analysis. On the academic front, the platform aims to advance the field of parallel computing and performance optimization. In research and innovation, it concentrates on overcoming key technological bottlenecks by providing core optimization solutions for domestic computing platforms, thereby enriching the domestic computing ecosystem. In terms of talent cultivation, the platform offers students invaluable opportunities to engage with cutting-edge industry challenges, enhancing their engineering, practical, and independent innovation capabilities. The platform's short-term goal is to continue deepening its project collaborations with industry leaders such as Huawei, ByteDance, and China Southern Power Grid, aiming to produce impactful technological achievements in parallel acceleration for existing heterogeneous platforms. Its long-term objective is to conduct systematic and forward-looking research centered on the three core areas mentioned above, with the ultimate aim of comprehensively enhancing China's independent innovation capabilities and core competitiveness in critical computing fields.
research direction：
1． Parallel Acceleration and Performance Optimization for Heterogeneous Computing This area focuses on the efficient utilization of domestic heterogeneous computing platforms, which are composed of units such as CPUs and GPUs. The research emphasizes maximizing hardware potential through advanced techniques including parallel algorithms, task scheduling, and compiler optimization. The primary objective is to overcome performance bottlenecks in compute-intensive applications like Artificial Intelligence and Big Data. Success in this area is critical for improving the overall performance and energy efficiency of domestic computing platforms and enhancing the nation's independent core competitiveness in the field of high-performance computing.
2． Software Reliability Assurance This area is dedicated to building a comprehensive quality assurance system that spans the entire software development lifecycle, employing methods such as automated testing, formal verification, and runtime monitoring. The research focuses on proactively identifying and rectifying software defects while predicting and mitigating systemic risks. This work is of paramount importance for ensuring the stable and secure operation of critical information infrastructure, such as power grids and financial systems, and serves as the foundation for establishing a trustworthy software ecosystem and reducing systemic risks.
3． User Behavior Analysis This area specializes in the in-depth analysis and modeling of user interaction data through the application of machine learning technologies. The research is centered on constructing accurate user profiles to understand usage patterns and predict user needs. The goal is to leverage data-driven insights to guide product optimization and innovation, thereby enhancing user experience, satisfaction, and engagement. This has a direct and profound impact on strengthening a product's market competitiveness and overall business value.
Talent team：
Team Overview:
The team consists of three professors, five associate professors and senior engineers, and one intermediate-level engineer.
Core Members:
Lu Lu Professor, School of Computer Science and Engineering, South China University of Technology. Professor Lu's primary research interests include software systems and architecture design, high-performance computing, and computing power optimization and parallel acceleration for heterogeneous platforms. He has published over 50 papers indexed by major citation databases and holds more than 30 patents (both granted and pending) and software copyrights. Professor Lu has led over 80 projects, including those funded by the National Natural Science Foundation of China (NSFC), sub-projects of the National Key R&D Program and National Science and Technology Major Projects, major applied science and technology projects of Guangdong Province and Guangzhou City, as well as numerous enterprise-commissioned development projects. As the lead researcher, he has received two Second Prizes of the Guangdong Provincial Science and Technology Progress Award.
Zhang Xinglin Professor, School of Computer Science and Engineering, South China University of Technology. Professor Zhang's research focuses on edge intelligence, federated learning, crowd intelligence, and AIoT. He has applied for over 20 invention patents, with more than 10 already granted. He has published over 70 academic papers, which have garnered more than 3,000 citations on Google Scholar. He has led or participated in 16 projects, including those funded by the NSFC, various provincial and municipal-level science and technology initiatives, and industry-funded projects.
He Junhui Associate Professor, School of Computer Science and Engineering, South China University of Technology. Associate Professor He's research primarily involves multimedia security and artificial intelligence security. He has led three projects under the General Program of the Natural Science Foundation of Guangdong Province, one Guangzhou Basic and Applied Basic Research Project, one provincial/municipal science and technology plan project, and several other projects. He has also served as a key member in a project funded by the NSFC-Guangdong Joint Fund and was appointed as a Guangdong Provincial Science and Technology Commissioner. He is a recipient of the Third Prize of the Guangdong Provincial Science and Technology Award, holds 7 granted invention patents, and has published over 20 academic papers.
Xu Lingling Associate Professor, School of Computer Science and Engineering, South China University of Technology. Associate Professor Xu's research areas include artificial intelligence security, data security and privacy protection, federated learning, and cryptography and its applications. She has published over 40 high-impact papers in prestigious international journals such as IEEE TPDS, IEEE TC, and IEEE TIFS. She has applied for more than 10 Chinese invention patents and has led over 10 projects, including those funded by the NSFC, provincial/ministerial-level grants, and enterprise commissions. She has also engaged in numerous research collaborations with universities and institutions both domestically and internationally.
Zhang Qin Associate Professor, School of Computer Science and Engineering, South China University of Technology. Associate Professor Zhang's research focuses on intelligent control technology and medical image processing. She has led 3 provincial/ministerial-level scientific research projects and 2 provincial/ministerial-level teaching and research projects, and has participated in numerous national or provincial/ministerial-level research projects. She has published nearly 30 papers, holds nearly 30 patents (invention and utility model) and software copyrights, has served as the chief editor for one textbook, and co-authored a monograph under the National 13th Five-Year Key Publication Plan.
Xian Jin Senior Engineer, School of Computer Science and Engineering, South China University of Technology. Mr. Xian's work focuses on research in artificial intelligence, big data, and image processing.
Projects：
1. Server RAID Card Project 2023
2. Ascend FFT Application Acceleration Cooperation
3. Building MLIR Expression Capabilities for Ascend
4. Research on GEMM Optimization Algorithms for Large Language Models
5. ACTLASS Collaboration Project
6. Software Development for Intelligent Recognition, Data Analytics, and Financial Business Modeling
7. Privacy-Preserving Methods in Crowdsensing
8. Key Technologies for Intelligent Edge Network Services in Federated Learning
9. Task Allocation Algorithms for Multi-Objective Crowdsensing
10. Online Incentive Mechanisms for Dynamic Users in Crowdsensing
Achievements：
Representative Achievements
Lu Lu
1. Yang, Zhanyu, Lu Lu, and Quanyi Zou.Ensemble Kernel-Mapping-Based Ranking Support Vector Machine forSoftware Defect Prediction. lEEE Transactions on Reliability(2024).
2. Guo, Yijie, Lu Lu, and Songxiang Zhu.Novel accelerated methods for convolution neural network with matrix core. The Journal of Supercomputing 79, no. 17 (2023):19547-19573.
3. Wang, Ruimin, Zhiwei Yang, Hao Xu, and Lu Lu. A high-performance batched matrix multiplication framework for gpus under unbalanced input distribution. The Journal of Supercomputing 78, no.2 (2022): 1741-1758.
4. Yang, Zhiwei, Lu Lu, and Ruimin Wang. A batched GEMM optimization framework for deep learning. The Journal of Supercomputing 78,no.11 (2022):13393-13408.
5. Hu, Yichang, Lu Lu, and Cuixu Li.Memory-accelerated parallel method for multidimensional fast fourier implementation on GPU. The Journal of Supercomputing 78, no.16 (2022):18189-18208.
Zhang Xinglin
1. F Tian, X Zhang, J Liang, Z Yang, Bidirectional service function chain embedding for interactive applications in mobile edge networks, IEEE TMC, 2024.
2. J Zhang, X Zhang, Multi-task allocation in mobile crowd sensing with mobility prediction, IEEE TMC, 2023.
3. X Li, X Zhang, Multi-task allocation under time constraints in mobile crowdsensing, lEEE TMC,2021.
4. L Luo, X Zhang, Federated Split Learning via Mutual Knowledge Distillation, IEEE TNSE, 2024.
5. Y. Zhang, X, Zhang, incentive Mechanism with Task Bundling for Mobile Crowd Sensing. ACM TOSN, 2023.
He Junhui
1. Junhui He, junxi chen, shaohua Tang. Reversible Data Hiding in JPEG lmages Based on Negative Influence Models, lEEE Transactions on Information Forensics and security,2020,15:2121-2133.
2. Junhui He, Junxi chen, Weigi Luo, shaohua Tang; and Jiwu Huang; A Novel High-Capacity Reversible Data Hiding scheme for Encrypted JPEG Bitstreams, lEEE Transactions on Circuits and Systems for Video Technology, 2019, 29(12):3501-3515.
3. Junhui He, shuhao Huang, Shaohua Tang, and liwu Huang, JPEG lmage Encryption with Improved Format Compatibility and File Size Preservation,lEEE Transactions on Multimedia, 2018, 20(10): 2645-2658.
4. Yingxuan Chen,Junhui He, yunting Xian. Reversible data hiding for JPEG images based on improved mapping and frequency ordering, signal Processing,2022,198:108604.
5. Junhui He, Yuzhang Xu, Weigi Luo, shaohua Tang, and jiwu Huang, A Novel Selective Encryption scheme for H.264/Avc Video with lmproved Visual Security, Signal Processing lmage Communication, 2020.
Xu Lingling
1. L. Xu, X. Chen, F. Zhang, etc. ASBKS: Towards attribute set based keyword search over encrypted personal health records. lEEE Transactions on Dependable and Secure computing, 2021.
2. L. Xu, W. Li, F. Zhang, etc. Authorized keyword searches on public key encrypted data with time-controlled keyword privacy, IEEE Transactions on Information Forensics and Security,2020.
3. L. Xu, Z. Sun, W. Li, etc. Delegatable searchable encryption with specified keywords for EHR systems. Wireless Networks, 2024.
4. J. Yao, L. Xu, Online/offline Attribute-based Boolean Keyword Search for Internet of Things. The Computer Journal, 2023.
Zhang Qin
Xian Jin
1. LPDA: Cross-Project Software Defect Prediction Approach via Locality Preserving and Distribution Alignment, International Journal of Advanced Computer Science & Applications, 2023.
2. FGSR: A Fine-Grained Ship Retrieval Dataset and Method in Smart Cities, Wireless Communications and Mobile Computing, 2022
Platform Construction ：
The laboratory currently has 60 square meters of space, equipped with 30 computers, 5 servers, and 4 high-performance computing graphics cards.
Cooperation and Communication：
The team has demonstrated highly effective industry-university-research partnerships through its collaboration on several key projects, including the Server RAID Card project with the Ministry of Industry and Information Technology, the Ascend Operator Development and Performance Optimization project with Huawei, the Artificial Intelligence Benchmarking and Operator Library Development project with China Southern Power Grid, and the Cutlass Template Library Development project with ByteDance. Further collaborations extend to the Key Technologies and Equipment for Intelligent Operational Safety Management and Emergency Support in Highway Tunnel Networks project with the Provincial Department of Science and Technology, and the Application of an Industrial Big Data Analysis and Collection Cloud Platform in the High-Salt, Dilute-State Soy Sauce Fermentation Process project with the Zhongshan Science and Technology Bureau, all of which underscore the team's impactful results in applied research and development.
Talent development：
In recent years, we have supervised 13 Ph.D. students—5 have graduated and 8 are currently enrolled—and 32 master’s students, of whom 17 have completed their degrees and 15 are still in progress.
Openness and Sharing：
l HPL-GPU: Optimized the HPL implementation for AMD platforms, achieving a 20% overall performance improvement. Contributed over 30,000 lines of source code, accounting for more than 95% of the total codebase.
l Assisted Pengcheng Laboratory in developing a single-node, continuous multi-core, multi-task acceleration operator, increasing computational efficiency from an initial 244 TFLOPS to 315 TFLOPS—representing a 29% performance gain.
l Developed an internet data collection and user behavior analysis platform, which has been successfully deployed across 20 enterprises and individual developers, including notable users such as Elite, Meiweixian, and Cloudy.
Contact Information：
Address: Building B3, School of Computer Science and Engineering, South China University of Technology, 382 Waihuan East Road, Higher Education Mega Center, Panyu District, Guangzhou, China
