On the morning of November 18, 2023, the "Computer Science Lecture" series of activities - Expert Academic Report Meeting, jointly organized by the School of Computer and Electronic Information of Guangxi University, the Graduate Student Association of the School of Computer and Electronic Information of Guangxi University, and the CCF Guangxi University Student Branch of the Chinese Computer Society, was held in the second conference room of Junwu Academy. This event invited Professor Hao Peng from Zhejiang Normal University, Professor Rui Mao from Shenzhen University, Professor Gang Wang from Nankai University, and Professor Jigang Wu from Guangdong University of Technology to hold a wonderful expert academic presentation on topics such as "Advanced Technologies and Applications of Big Model Security", "Big Data Architecture: General Data Processing Mode for Measuring Space", "Research Progress on Hard Disk Fault Prediction in Data Centers", and "Software and Hardware Optimization Technology for Neural Network Computing". Professor Cheng Zhong from the college presided over the report meeting, and about 100 graduate students from the college attended.
Figure 1: Professor Rui Mao (first from left), Professor Gang Wang (second from left), Professor Cheng Zhong (third from left), Professor Jigang Wu (fourth from left), Professor Hao Peng (fifth from left)
At the beginning of the conference, Professor Jigang Wu reported on "Software and Hardware Optimization Techniques for Neural Network Computing" highlighted that model training and result inference have become key guarantee links for intelligent service quality. The report focuses on the application and computing characteristics of neural networks. It introduces optimization techniques for service quality assurance, hardware accelerator design for neural network computing, and memory based integrated acceleration technology from different levels of software, hardware, and software hardware integration. It explores and shares software optimization methods for hardware architecture, reports current research progress, and exchanges the latest research results.
Figure 2: Professor Jigang Wu giving a presentation
Next, Professor Gang Wang introduced the research progress of the Parallel and Distributed Software Research Laboratory at Nankai University in using recursive neural networks, progressive gradient regression trees, variational autoencoders and stream models, and artificial intelligence large models to construct high-performance computing system hard disk fault prediction models and design algorithms, in order to significantly improve detection rates, reduce false alarm rates, and solve the problem of significant cross model performance degradation of prediction models.
Figure 3: Professor Gang Wang giving a presentation
Next, Professor Hao Peng will introduce the development history, security risks, and solutions of big model technology, as well as the application of big models in the field of network security. He will review the evolution from early simple models to today's complex big models and emphasize their importance in handling large-scale data and complex tasks; Focus on discussing the security risks of large models such as data privacy breaches, model deception, and bias, and explore solutions including differential privacy, adversarial training, and fairness algorithms; Showcase the application of large models in threat detection, intrusion prevention, and security policy automation, highlighting their potential in enhancing network security defense capabilities.
Figure 4: Professor Hao Peng giving a presentation
Finally, Professor Rui Mao presented the topic of "Big Data Generalization: A General Data Processing Model for Metric Spaces". General data processing techniques have been favored by commercial data processing systems due to their wide applicability and relatively low average development and maintenance costs. The more versatile the information, the less information can be utilized, so there is often a contradiction between universality and performance. General systems based on numerical types (scalars, vectors, matrices, tensors, etc. ) have high performance but low generality, while general systems based on graphs have high generality but low performance. Big data generalization abstracts data into points in a metric space without requiring the internal structure of the data. It only utilizes the triangulation of distances between data for data processing. Its generality and performance are between those of digital type based and graph based general-purpose systems, achieving a compromise between generality and performance. It is expected to become an important component of the next generation of general-purpose data processing systems.
Figure 5: Professor Rui Mao giving a presentation
The wonderful presentation of this academic conference has expanded students' academic horizons and helped them clarify their research ideas. The students have benefited greatly, like basking in the spring breeze, which will surely inspire our faculty and students to work hard, forge ahead, and innovate.
