Prof. Jihua Zhu

Xi'an Jiaotong University, China

Jihua Zhu received the Ph.D. degree in pattern recognition and intelligence systems from Xi'an Jiaotong University, Xi'an, China, in 2011. He is currently a Professor with the School of Software Engineering, Xi'an Jiaotong University, China. He is senior member of IEEE and Associate Editor of EURASIP Journal on Image and Video Processing. His research interests include computer vision and machine learning.

Speech Title: Self-Supervised Learning and Efficient Transfer for 3D–4D Point Cloud Understanding
演讲题目：面向3D—4D点云理解的自监督学习与高效迁移方法
Abstract: Leveraging large-scale unlabeled point clouds for generalizable representations and transferring static 3D knowledge to dynamic 4D scenes are key directions in 3D vision. Among self-supervised methods, masked point modeling is notable for its simplicity and strong transferability, yet faces three challenges: fixed masking fails to adapt to local geometry, deterministic reconstruction cannot capture multiple plausible structures of missing regions, and static-to-dynamic transfer suffers from modality gaps, few-shot overfitting, and high resource costs. This talk presents our recent progress on enhancing self-supervised point cloud representations and efficient transfer, covering representation complementarity, adaptive masking, probabilistic reconstruction, static–dynamic feature alignment, and lightweight spatiotemporal adaptation, toward unified 3D–4D point cloud understanding.
摘要：利用大规模无标签点云学习通用表征，并将静态3D知识迁移至动态4D场景，是三维视觉领域的重要研究方向。现有自监督方法中，掩码点建模因结构简单、迁移性强而备受关注，但仍面临三方面挑战：固定掩码策略难以适配局部几何差异，确定性重建无法刻画缺失区域的多重可能结构，静态模型向动态数据迁移时存在模态差异、小样本过拟合及资源开销大等问题。本次报告将介绍我们在点云自监督表征增强与高效迁移方面的近期研究进展，重点围绕表征互补性构建、自适应掩码策略设计、概率化重建框架、静态—动态特征对齐，以及轻量化时空适配机制等若干方向展开讨论，旨在为3D与4D点云理解的统一建模提供有益的探索与参考。

Prof. Kaibing Zhang

Xi'an Polytechnic University, China

Kaibing Zhang received the M.Sc. degree in computer software and theory from Xihua University, Chengdu, China, in 2005, and the Ph.D. degree in pattern recognition and intelligent system from Xidian University, Xi’an, China, in 2012. He is currently a Professor at the School of Computer Science, Xi’an Polytechnic University, Xi’an. His main research interests include pattern recognition, computer vision, and image super-resolution reconstruction. In these areas, he has published around 50 technical articles in refereed journals and proceedings, including IEEE Transactions on Image Processing, IEEE Transactions on Neural Networks and Learning Systems, IEEE Transactions on Circuits and Systems for Video Technology, IEEE Signal Processing Letters, Knowledge-Based Systems, Signal Processing (Elsevier), Neurocomputing, CVPR, and ICIP.

Speech Title: Research and Advances in Multimodal Knowledge-Driven Controllable Fashion Garment Image Generation
演讲题目：多模态知识驱动的时尚服装图像可控生成研究与进展
Abstract: In recent years, generative artificial intelligence (GenAI), represented by diffusion models and multimodal large language models, has advanced rapidly, driving image generation from content creation toward multimodal understanding, knowledge-enhanced generation, and fine-grained controllability. This evolution provides a new technological paradigm for the digital and intelligent transformation of fashion design. Fashion garment images involve complex structural, textural, and semantic attributes, while multimodal generation still faces significant challenges in semantic alignment, structure–texture coordination, local attribute control, and appearance consistency. This report first reviews the development and evolution of GenAI for image generation and analyzes the key challenges in multimodal controllable fashion garment image generation. It then presents our research on retrieval-augmented multi-granularity feature-aware generation and multimodal-conditioned fine-grained garment image generation, followed by recent studies on local multi-attribute editing and pose-constrained garment generation. Finally, the report discusses future directions, including multimodal knowledge enhancement, local multi-attribute editing, and physics-prior-driven garment generation, with the aim of advancing fashion image generation from visual content creation toward high-fidelity intelligent design that integrates semantic, structural, material, and physical knowledge. These developments are expected to provide new perspectives for leveraging AI in digital fashion and intelligent garment design.
摘要：近年来，以扩散模型和多模态大模型为代表的生成式人工智能快速发展，推动图像生成由内容创作向多模态理解、知识增强与精细化可控生成演进，为时尚服装数字化与智能化设计提供了新的技术范式。时尚服装图像具有复杂的结构、纹理与语义属性，多模态条件下的语义对齐、结构—纹理协同、局部属性控制及外观一致性仍面临诸多挑战。 本报告首先介绍生成式人工智能在图像生成领域的发展与演进，分析多模态时尚服装图像可控生成面临的关键问题；随后介绍本课题组围绕检索增强的多粒度特征感知生成和多模态条件驱动的细粒度服装图像生成开展的研究，进一步探讨局部多属性编辑与人体姿态约束下的服装生成方法。最后，报告将围绕多模态知识增强、局部多属性编辑以及物理先验驱动的服装生成等方向展望未来发展趋势，探索时尚服装图像生成由视觉内容创作向融合语义、结构、材质与物理规律的高保真智能设计演进，为人工智能赋能数字时尚与智能服装设计提供新的思路。

Invited Speakers

Prof. Mohan Lal KOLHE
University of Agder, Norway

Prof. Dr. Mohan Lal Kolhe is a distinguished academic and researcher, currently serving as a Full Professor of Smart and Sustainable Electrical Energy Systems with Renewables and Hydrogen at the University of Agder in Norway. He is an internationally recognized expert in smart grids, renewable energy technologies, and hydrogen systems, with over three decades of global academic leadership and innovation.
Prof. Kolhe’s research spans integrated renewable energy systems, power electronics and grid stability, hydrogen production and utilization, and AI-enabled energy management. He has held significant academic positions at world-renowned institutions, including University College London (UK/Australia), University of Dundee (UK), University of Jyvaskyla (Finland), and the Hydrogen Research Institute (Canada). He also served on the Government of South Australia’s first Renewable Energy Board (2009–2011), contributing to progressive renewable energy policy development.
A prolific author and editor, Prof. Kolhe has published extensively and secured competitive funding from leading research bodies such as the Norwegian Research Council, EU, EPSRC, and BBSRC. His work has been ranked among the top 2% of scientists globally by Stanford University metrics, reflecting its high impact and influence in the transition toward sustainable energy systems.

Assoc. Prof. Chao Fang
Beijing University of Technology, China

Chao Fang received his B.S degree in Information Engineering from Wuhan University of Technology, Wuhan, China, in 2009, and the Ph.D. degree with the State Key Laboratory of Networking and Switching Technology in Information and Communication Engineering from Beijing University of Posts and Te4lecommunications, Beijing, China, in 2015. He joined the Beijing University of Technology in 2016 and now is an associate professor. From August 2013 to August 2014, he had been funded by China Scholarship Council to visit Carleton University, Ottawa, ON, Canada, as a joint doctorate. Moreover, he is the visiting scholars of University of Technology Sydney, Commonwealth Scientific and Industrial Research Organization, Hong Kong Polytechnic University, Kyoto University, Muroran Institute of Technology, and Queen Mary University of London. Dr. Fang is the senior member of IEEE, and the vice chair of technical affairs committee in IEEE ComSoc Asia/Pacific Region (2022-2023). Moreover, he served as the Technical Program Committee Chair of SPCNC 2024, the Session Chairs of ICC 2015, ICCC 2023, and WCNC 2024, Workshop Chairs of ICFEICT (2022-2024) and ICNCIC (2023-2024), and Poster Co-Chair of HotICN 2018. He won the Best Paper Award of IEEE ICFEICT 2022 and 2024, ICCSN 2024, and NCIC 2024. His current research interests include intelligent analysis and control, and intelligent cloud-edge-terminal cooperation computing.

Speech Title: Collaborative Allocation and Intelligent Optimization of Service-Driven Cloud Radio Access Network Resources
Abstract: In order to meet the service requirements of the emerging applications such as extended reality, 8K ultra-high definition video transmission and industrial Internet of Things in terms of massive user access, heterogeneous mobile traffic processing, ultra-low latency, ultra-high reliability and other aspects, cloud-edge collaboration, as the core of cloud radio access networks (C-RAN), has been increasingly concerned and risen to the height of national development strategy. At present, the problem on cooperative allocation and optimization of cloud-edge-end resources in C-RAN is still in the initial research stage, lacking systematic and in-depth research, which makes it difficult to adaptively guarantee the differentiated service requirements of network business. Therefore, by sorting out and referring to the research ideas and methods related to cloud computing and fog computing, and drawing on future network concepts such as "separation of control and forwarding" in software-defined networking and "in-network caching" in information-centric networking, the project focuses on collaborative allocation and intelligent optimization mechanisms of service-driven C-RAN resources from the perspective of cross-layer and cross-domain cooperation. To improve the overall service capacity and satisfy the differentiated service requirements of massive applications, key technologies such as multi-user-oriented cross-layer collaboration allocation and intelligent optimization of cloud-edge-terminal resources, multi-business-oriented cross-layer collaboration and intelligent resource allocation, multi-business-oriented cross-domain collaboration and intelligent resource allocation will be solved in C-RAN environments, providing customized service for network applications.

Assoc. Prof. Gang Yang
Information Support Force Engineering University, China

Gang Yang, Information Support Force Engineering University, China Title: Dataflow-Driven Decomposition for Reliable Fuzz Harness Generation Abstract: Automated fuzz harness generation is critical for effective gray-box fuzzing of C projects, yet existing LLM-based methods suffer from imprecise function selection and error-prone monolithic generation. This talk presents SynapseFlow, a framework that addresses these challenges through two core mechanisms. First, it performs dataflow-aware function aggregation by constructing a structural flow graph over source code, classifying functions into input stream, process, and helper roles, and extracting minimal function triplets that capture coherent dataflow. Second, it synthesizes harnesses via a decomposed four-stage workflow—function documentation, structure snippet stitching, rough code assembly, and harness optimization—governed by a staged rollback algorithm that detects and recovers from hallucinations. Evaluated on 25 real-world C projects, SynapseFlow achieves up to 3.07× higher branch coverage than state-of-the-art tools and has discovered 7 previously unreported bugs. This talk will focus on the methodological design, implementation insights, and practical lessons.

Speech Title: Dataflow-Driven Decomposition for Reliable Fuzz Harness Generation
Abstract: This talk introduces SynapseFlow, a framework for automatic fuzz harness generation in C projects. It combines dataflow-aware function aggregation—using structural flow graphs and function triplets—with a staged, rollback-enabled LLM workflow to mitigate hallucinations. Experiments on 25 real-world projects show significant coverage improvements and 7 new bug discoveries, with a focus on the underlying method.

Assoc. Prof. Chunlei Xia
Yantai Institute of Coastal Zone Research,Chinese Academy of Sciences, China

Chunlei Xia received his Ph.D. in the Department of Electronics and Electrical Engineering from Pusan National University in 2012. He then served as a Senior Researcher at the SPENALO Robot Center, PNU, before joining the Yantai Institute of Coastal Zone Research, Chinese Academy of Sciences (CAS) in 2013. His research interests include marine monitoring technology, underwater optical imaging, intelligent ocean-observation instruments, and marine organism imaging and recognition systems. He has undertaken more than ten national-level research projects, including grants from the National Key R&D Program of China, the National Natural Science Foundation of China (NSFC), and talent programs of CAS. He has published over 40 papers, obtained 6 patents, and served as Guest Editor and reviewer for several SCI-indexed journals.

Speech Title: AI-Enabled Marine Ecological and Environmental Sensing Technologies and Applications
Abstract: As a key component of marine ecosystems, marine organisms play a critical role in monitoring environmental change, safeguarding marine ecological security, and managing aquaculture. Conventional manual monitoring approaches suffer from low timeliness, high cost, and insufficient data continuity, and thus struggle to meet the demands of large-scale, long-term marine ecological monitoring. To address this, the present work targets practical application scenarios in marine ecological monitoring and develops intelligent, automated techniques for the online monitoring of marine organisms. By constructing monitoring systems operating at multiple scales, real-time acquisition and analysis of the population structure, spatiotemporal distribution, and behavioral dynamics of target organisms are achieved. The systems offer high timeliness, non-invasiveness, and high accuracy, providing reliable data support and technical assurance for marine ecological assessment, aquaculture management, and early warning of ecological disasters.

Prof. Dongshik Kang
University of the Ryukyus, Japan

Dongshik Kang is a Professor in the Department of Computer Science & Intelligent Systems, Faculty of Engineering, at the University of the Ryukyus, Okinawa, Japan. He received his Ph.D. in Electrical and Information Science from Osaka Prefecture University, Japan, in 1999. He joined the University of the Ryukyus as an Assistant Professor in the Department of Information Engineering in April 1999 and has since contributed to education and research in intelligent information systems. Alongside his academic career, he has actively collaborated with industry and research institutions in Japan, promoting the practical application of intelligent technologies. His research focuses on machine learning, intelligent systems, biological data analysis, educational technology, and artificial intelligence, with particular interests in applying AI to education, and environmental data analysis. Professor Kang is a Fellow of the Institute of Electronics, Information and Communication Engineers (IEICE), a member of the Information Processing Society of Japan (IPSJ), and currently serves as Chair of the Okinawa Branch of the Institute of Internet, Broadcasting and Communication (IIBC) in Japan.

Speech Title: Estimation of Student’s Interest Level in Classroom Video
Abstract: Recent educational reforms have emphasized personalized and collaborative learning; however, these approaches have also increased teachers’ workloads, creating a need for efficient classroom evaluation methods. To support teachers in reviewing classroom activities, this study proposes a student interest level estimation system that analyzes classroom videos and automatically estimates students’ engagement levels. The proposed system utilizes OpenPose to extract human pose in- formation and generates temporal behavioral sequences based on facial orientation, facial motion, and hand gestures. Student interest levels are then estimated from these behavioral patterns. To improve the performance of a previously developed system, two major enhancements are introduced. First, a k-means-based human tracking method is employed to improve student detection and identification using spatial information. Second, L2CS-Net, a fine-grained gaze estimation model, is incorporated to improve face orientation detection through accurate estimation of yaw and pitch angles. By integrating robust human tracking and gaze estimation techniques, the proposed system aims to provide a more reliable visualization of student engagement in classroom environments. The effectiveness of the proposed approach is evaluated using real classroom videos, demonstrating its potential as a practical tool for supporting educational assessment and classroom reflection.

Abhradeep Chatterjee
NTT DATA Services, USA

Abhradeep Chatterjee is an Associate Director at NTT DATA Services, USA, with more than 15 years of experience in enterprise technology leadership, AI-driven IT operations, digital transformation, cloud operations, and resilient production systems. His work focuses on helping large organizations modernize operational environments by applying predictive analytics, observability, intelligent automation, governance controls, and production-scale AI capabilities. He has led enterprise initiatives involving incident management, operational intelligence, automation, system reliability, and value-driven transformation across complex business and technology environments. He is an IEEE Senior Member and MIET, and actively contributes to the global technology community as an invited speaker, technical program committee member, reviewer, and international judge. His research and professional contributions span AI-enabled operations, autonomous remediation, enterprise resilience, trustworthy AI, digital transformation, and decision-support systems. He has authored more than 30 publications and has received significant scholarly citations for his work in AI, operational intelligence, and distributed systems.

Speech Title: Trust-Aware Autonomous Operations for Resilient Enterprise Computer Systems
Abstract: Modern computer systems are increasingly expected to operate across distributed, cloud-enabled, and business-critical environments where downtime, delayed response, and uncontrolled automation can create significant organizational impact. As enterprises adopt AI-driven operations and autonomous remediation, the key challenge is no longer only whether systems can detect and respond to issues, but whether they can do so reliably, transparently, and within well-defined control boundaries. This invited speech will discuss how trust-aware autonomous operations can strengthen the resilience of enterprise computer systems. The talk will examine the evolution from reactive monitoring to predictive operational intelligence, including the role of observability, event correlation, anomaly detection, intelligent automation, human-in-the-loop governance, rollback mechanisms, and policy-aware decisioning. It will also highlight practical safeguards such as confidence thresholds, escalation paths, auditability, and blast-radius control that help prevent automated actions from amplifying failures. Drawing on enterprise-scale production experience, the session will provide a pragmatic framework for designing computer systems that are not only intelligent, but also resilient, explainable, and operationally accountable.