Prof. William YeohHong Kong Metropolitan University, China Speech Title: Human-Centric Cybersecurity: My Journey and Insights. Abstract: Zero trust cybersecurity is beginning to replace traditional perimeter-based security strategies and is being adopted by organizations across a wide range of industries. However, the implementation of zero trust is a complex undertaking, different from traditional perimeter-based security, and requires a fresh approach in terms of its management. As such, a clear set of critical success factors (CSFs) will help organizations to better plan, assess, and manage their zero trust cybersecurity. In response, we investigated the CSFs for implementing zero trust cybersecurity by conducting a three-round Delphi study to obtain the consensus from a panel of 12 cybersecurity experts. We built a multi-dimensional CSFs framework that comprises eight dimensions, namely identity, endpoint, application and workload, data, network, infrastructure, visibility and analytics, and automation and orchestration. Based on the CSFs, we developed a maturity assessment framework enabling organizations to evaluate their zero trust maturity. This paper contributes to a theoretical understanding of how to deploy zero trust from multiple dimensions and offers a viable guidance framework for organizations from a practical perspective. This paper is useful for organizational stakeholders who are in the process of planning, reviewing, or implementing zero trust cybersecurity. |
Prof. Wenke ZangShandong Normal University Speech Title:Research on Adaptive Optimization of Density Peak Clustering Algorithms for Complex Data. Abstract: Density Peaks Clustering (DPC) is a well-known clustering technique in the data mining field with fewer parameters as well as no iteration. However, DPC struggles to demonstrate superior clustering performance when processing complex datasets (e.g., manifold, multi-peak, or imbalanced data) owing to its reliance on subjective manual center selection, traditional local density and relative distance estimation methods, and incorrect allocation of data points. This report first provides an overview of the fundamental principles, research status, and core challenges of the DPC algorithm. Simultaneously, to address these limitations, this report proposes adaptive optimization approaches for various complex datasets from four perspectives: multi-feature-point representation, density voting and neighborhood diffusion, connectivity subgraphs construction, and optimization of density and distance estimation. Specifically, our research work encompasses: the development of a multi-feature-point density peak clustering algorithm adapted to cluster structures; the enhancement of cluster center identification and core cluster structure recognition based on density voting and neighborhood diffusion; the design of connectivity subgraphs constrution using superior nodes and fuzzy correlation to eliminate the need for manual center selection; the proposal of a density peak clustering algorithm with connected local density and punished relative distance, tailored for datasets with specific geometric shapes. This presentation will present a comparative analysis of these improved algorithms in terms of their core concepts, technical contributions, and experimental performance. Building upon the review of DPC algorithms, we will discuss their practical application potential and challenges. Experimental results demonstrate that these enhanced algorithms exhibit superior performance on synthetic datasets, real-world datasets, and image data, offering valuable insights and references for the future development of density peak clustering algorithms in complex data mining tasks. |
Prof. Shuai LiuHunan Normal University Speech Title: AI-assisted Class Diagnosis based on Behavioral Sequence. Abstract: AI-assisted class diagnosis based on behavioral sequence is an important research direction intheintegration of AI/GAI in education (AIED),mainly introduces techniques and methods inAI/GAI to analyze surface and fragmented behavioral data, thus revealing the deep and essential class information hidden behind the behavioral data. AI-assisted class diagnosis based on large-scale behavioral coding is a research hotpot. Manual coding for large-scale videos is impossible, and the existing high-similarity behaviors and abstract coding rules are not adaptable to automated coding with AI. Since the diversity of in-class behaviors, the existing method of behavioral diagnosis mainly focused on the classroom overall analysis and lacked process analysis, which made it difficult to diagnose process problems in the in-class teaching. Therefore, common behaviors in high-quality classes is mined for diagnosing in-class behaviors to solve these problems. Firstly, a new in-class behavior coding system for AI (AI-IBCS) is required to replace the high-similarity behaviors and abstract coding rules for AI-assited behavior recognition in large-scale behavior coding. Then, a near-optimal common subsequence mines common behaviors for class diagnose. Finally, 57 pre-service teachers is used as an example to conduct a diagnostic experiment for verifying the effectiveness of the proposed methods. Results of the consistency between manual and AI coding demonstrate that AI-IBCS can effectively coding in-class features. Since common behaviors differ across subjects, the proposed diagnostic results are more specific and efficient by comparing with experts' observations. |