CCID 2025 provides a premier venue for the presentation and discussion of research in the design, development,
deployment and evaluation of cloud computing and intelligent driving. It bridges together academic researchers
and industrial practitioners to share and exchange the latest developments in the area of high-performance
computing, autonomous driving, Large Language Models (LLM), and Embodied artificial intelligence (AI). We hope
that this conference will stimulate our participants to explore innovative advances and applications in
computing science and AI.
We gratefully acknowledge the support from University of Macau, Macau Science and Technology Development Fund
(FDCT) and Ministry of Science and Technology of China. We wish to express our heartfelt appreciation to the
keynoter speakers, invited speakers, students and volunteers for their help. We wish all participants and
colleagues a very pleasant stay in Macau.
Prof. Cheng-Zhong Xu, Chair Professor, University of Macau
Prof. Jianbing Shen, Professor, University of Macau
Prof. Carlos Jorge Ferreira Silvestre, Professor, University of Macau
09:00 - 09:10
Prof. Cheng-Zhong Xu, University of Macau
09:10 - 09:30
09:30 - 09:45
09:45 - 10:00
10:00 - 10:40
Prof. José Santos-Victor, University of Lisbon
10:40 - 11:10
11:00 - 11:30
Prof. Qi Hao, Sustech
11:30 - 12:00
Prof. Kejiang Ye, Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences
12:30 - 14:30
14:30 - 15:00
Prof. Shuai Wang, Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences
15:00 - 15:30
Prof. Yubin Zhao, SYSU
15:30 - 16:00
16:00 - 16:30
Prof. Juanjuan Zhao, Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences
16:30 - 17:00
Prof. Huanle Xu, UM
17:00 - 17:30
Prof. Minxian Xu, Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences
09:00 - 09:30
09:30 - 10:30
Prof. Nitesh Chawla, University of Notre Dame
10:30 - 11:00
11:00 - 11:30
Prof. Carlos Silvestre, UM
11:30 - 12:00
Prof. Xitong Gao, SIAT
12:30 - 13:30
13:30 - 14:30
Prof. Guoliang Li, Tsinghua University
15:00 - 15:30
Prof. Joel Reis, UM
15:30 - 16:00
Dr. Nuno Ferreira Duarte, University of Lisbon
16:00 - 16:30
Abstract: I will present an overview of recent developments and emerging challenges in Robotics and
Artificial Intelligence, with particular emphasis on the use of artificial vision and humanoid robotics to
advance our understanding of key functions of the human brain.
The talk will focus on the problem of non-verbal communication, a crucial component of human–human
interaction, and action understanding. I will discuss the human ability to anticipate the actions and
intentions of others through the interpretation of subtle body, arm, head, and eye movements. Building on
specially designed human experiments, we propose a computational model that captures the interplay between
non-verbal cues during handover interaction tasks.
Throughout the presentation, I will illustrate our work with examples drawn from the humanoid robotic
platforms we have developed and used over the years: from our first robot, Baltazar, to iCub—the
open-source, highly sophisticated humanoid we co-designed—and Vizzy, our socially assistive robot. These
platforms serve both as experimental testbeds to study human cognition, learning, and sensorimotor
coordination, and as vehicles for exploring new engineering approaches to artificial systems. Finally, I
will provide an overview of several applications of visual action understanding, highlighting their
relevance for robotics, human–machine interaction, and cognitive science.
Biography: José Santos-Victor is a Distinguished Professor at Instituto Superior Técnico (IST), University
of Lisbon, in the Department of Electrical and Computer Engineering. He is a senior researcher at the
Institute for Systems and Robotics | Lisboa (ISR-Lisboa) and the founder of VisLab – the Computer and
Robot Vision Laboratory. He also serves as the Coordinator of LARSyS – the Laboratory of Robotics and
Engineering Systems, a large multidisciplinary research centre.
His research interests focus on computer and robot vision, as well as biologically inspired models of
human cognition, including the development of artificial models using humanoid robots and human–robot
interaction (HRI). He has supervised 25 Ph.D. students and has been the scientific lead for ISR-Lisboa in
more than ten large European projects in Computer Vision and Robotics (including MIRROR, CONTACT, ROBOSOM,
ROBOTCUB, FIRST-MM, POETICON+), and is currently involved in the EU FET-Pathfinder project REPAIR: AI and
Robotics Meet Cultural Heritage.
He has held several leadership roles, including Chair of the Department of Electrical and Computer
Engineering (2021–24), IST Vice-President for International Affairs (2006–14), Secretary-General of the
CLUSTER Network of leading European universities of technology (2010–12), and President of ISR-Lisboa
(2015–22).
Professor Santos-Victor is a Fellow of the IEEE, a Fellow of ELLIS – the European Laboratory for Learning
and Intelligent Systems, and a Corresponding Member of the Lisbon Academy of Sciences.
Abstract: This talk will introduce our recent research efforts and results in dataset, simulation and integrated validation platform for developing trustworthy AD systems. We have optimized the AD dataset from statistical and perception uncertainty perspectives, leading to efficient and effective training and testing datasets. We have explored multiple simulation approaches including surfel, mesh, NeRF, 3D Gauss and VLM for long-tail and safety-critical scenario generation. Besides, sensor simulation models for camera and LiDAR as well as different sim-to-real schemes are also investigated to generate high-fidelity sensory data. The scheme of hazard factors injection and the hardware-in-loop testing platform are developed as an integrated high-efficiency testing framework for trustworthy AD system development.
Biography: Professor in the Dept. of Computer Science and Engineering at Southern University of Science and Technology，PhD from Duke University, Executive Director of the Research Institute of Trustworthy Autonomous Systems and the Deputy Director of Shenzhen Key Lab for Robot Vision and Navigation. His group research is sponsored by NSFC, Intel, Huawei and other agencies. Published more than 130 papers in the related areas.
Abstract:
Low altitude economy is a strategic emerging industry direction for the future. According to the
prediction of the Civil Aviation Administration of China, the market size of China's low altitude economy
will be about 1.5 trillion yuan by 2025, and will reach 3.5 trillion yuan by 2035. In December 2024, the
National Development and Reform Commission established the Low Altitude Economic Development Department to
coordinate and promote the development of China's low altitude economy. However, the current low altitude
computing system in China is still in its initial development stage, facing a series of challenges such as
data, models, algorithms, systems, and computing power.
This report first systematically introduces the background of low altitude economy and introduces the
national key research and development plan project "Research on Low Altitude Intelligent Computing Theory
and Method Based on LLM". This project will empower low altitude computing with LLM technology, conduct
research on the "state perception, data fusion, task scheduling, resource allocation, planning and
decision-making" involved in low altitude intelligent computing, and propose a new theory and method of
low altitude intelligent computing based on LLM. Finally, will introduce the Yunyuan - Low Altitude Large
Model and related work recently developed by the project team.
Biography: Kejiang Ye is currently a Professor and the Director of the Research Center for Cloud Computing, Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences. He received his B.S and Ph.D degree both from Zhejiang University and was a Post Doctoral Research Associate at Carnegie Mellon University (CMU). His research interests include Cloud Computing, Big Data and Industrial Internet. He is a Senior Member of IEEE and a Distinguished Member of China Computer Federation (CCF).
Abstract: Next generation robot navigation should be accurate, real-time, efficient, generalized, and user-friendly. This appears to be difficult for a standalone model. This talk will discuss how to leverage multi-model fusion to meet these requirements. First, we will discuss scene representation models incorporating geometric, semantic, motion, and electromagnetic information. Second, we will present a multi-scale planning system, consisting of global cognition, regional optimization, and real-time execution models, enabling global trajectory generation, regional path adjustment, and local obstacle avoidance. Finally, we demonstrate how multi-model fusion enhances robot navigation performance in complex environments.
Biography: Dr. Shuai Wang received the Ph.D. degree from The University of Hong Kong. He is currently an Associate Professor with Shenzhen Institutes of Advanced Technology (SIAT), Chinese Academy of Sciences. His research interests are in connected autonomous systems. He has published over 100 papers in related journals and conferences, including TRO, TIT, ICRA, IROS, and has received three IEEE Best Paper Awards. He currently serves as an Associate Editor for IEEE Transactions on Mobile Computing and IEEE Open Journal of the Communications Society.
Abstract: Wireless sensing is emerging as a rapidly growing research area, with a wide range of enabling technologies, including millimeter-wave radar, WiFi, and other radio-frequency based approaches. The widespread availability and cost effectiveness of WiFi enables ubiquitous sensing in daily life. With the rapid development of AI technology, wireless sensing can be extend to a wide range of applications, e.g., smart home, smart healthcare, and smart factory. Typical WiFi-based applications attain more attention due to the wide deployment. Within WiFi, channel state information (CSI) provides various signals which are highly correlated with the sensing applications. In this talk, we will have a brief review of the recent progress of WiFi based wireless sensing. The data-driven and model-driven methods are introduced. In addition to the algorithms, we also discuss about how to implement wireless sensing system for practical usages. The future research directions and challenges are presented as well.
Biography: Yubin Zhao is the IEEE senior member. He received his B.S. and M.S. in 2007 and 2010 respectively from Beijing University of Posts and Telecommunications (BUPT), Beijing, China. He received his Ph.D degree in computer science in 2014 from Freie Universitӓt Berlin (FU Berlin), Berlin, Germany. He joined Center for Cloud Computing as an associate professor, Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China, in 2014. He is now the associate professor in School of Microelectronics Science and Technology, Sun Yat-Sen University, Zhuhai, China. He serves as the guest editor and reviewer for several journals. His current research interest includes wireless power transfer, indoor localization and target tracking. He was a recipient of IEEE distinguished service award in IEEE SmartData 2023 and the Excellent Teacher Award of College Computing Science in China, 2023. He serves as the vice technical chair in IEEE SmartData 2023, IEEE ScalCom 2022, publicity chair of IEEE NFV-SDN 2019 and tutorial chair of IEEE NFV-SDN 2020.
Abstract: This talk systematically reviews our team’s research in transportation AI over the past decade, with a focus on core tasks including traffic data cleaning, traffic forecasting, and travel behavior analysis. While consistent progress has been made in optimizing traditional models, fundamental challenges—such as barriers to integrating multi-source heterogeneous data, inadequate robustness in environmental perception, and limited cognitive reasoning in complex scenarios—continue to impede substantial performance gains. Recent breakthroughs in multimodal foundation models offer a new paradigm for addressing these long-standing issues. Their strong cross-modal alignment capabilities, rich embedded world knowledge, and deep contextual reasoning mechanisms are proving instrumental in deciphering the complexity of transportation systems. In this context, the talk examines how multimodal foundation models can provide transformative solutions to these persistent challenges in transportation AI, and reports our team’s preliminary work on traffic data augmentation, alignment of multimodal traffic data, and the development of LLM-based intelligent agents for traffic decision-making.
Biography: Juanjuan Zhao is an Associate Researcher and PhD Supervisor at the Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences. She has long been dedicated to applied research at the intersection of artificial intelligence, big data, and domains such as smart cities, traffic management, and public safety. Her core research interests include big data quality assessment, multimodal spatio-temporal foundation models, multimodal intelligent agents, and trajectory privacy protection. Leveraging years of large-scale traffic data from Shenzhen, her team developed intelligent traffic perception and reasoning technologies that were awarded the Second Prize of the Guangdong Province and Shenzhen Municipal Science and Technology Invention Award in 2022. These technologies have been deployed by enterprises and government agencies including ZTE Corporation, Shenzhen Bus Group, and Shenzhen Metro, and have led to the incubation of a national high-tech enterprise. Dr. Zhao has served as principal investigator or co-investigator on seven national research projects, including grants from the National Natural Science Foundation of China (NSFC) and sub-projects of the National Key R&D Program, and has participated in over 30 national, provincial, and municipal projects. She is also a peer reviewer for NSFC and science and technology programs in Guangdong Province and Shenzhen. She has published more than 60 papers in prestigious venues such as CVPR, ICCV, ACM SIGSPATIAL, and IEEE Transactions on Intelligent Transportation Systems (TITS), and holds 30 granted patents, five of which have been successfully commercialized.
Abstract: This talk is structured into two parts, each introducing a novel mechanism aimed at optimizing LLM serving in cloud environments. First, we present a new memory management strategy designed to address memory bottlenecks. The key innovation lies in an adaptive KV cache storage method, which selectively retains critical tokens while dynamically recomputing non-cached tokens in parallel with the decoding process. This approach strikes a balance between memory usage and computational overhead, enhancing service throughput while maintaining decoding SLOs. Second, we introduce Attention Piggybacking, a mechanism that leverages both CPU and GPU resources to improve throughput for best-effort (BE) services and ensure SLOs for latency-sensitive (LS) services. By offloading the Attention computation for BE services to CPUs on-the-fly, it alleviates pressure on GPU resources. At the heart of this design is an asynchronous communication protocol between CPU and GPU streams, which prevents GPU blocking during the aggregation of Attention results, thereby maximizing hardware utilization.
Biography: Huanle Xu is currently an assistant professor in the Department of Computer and Information Science, University of Macau. He is also the director of the Cloud Computing and Distributed System Lab. Huanle received his bachelor degree from Shanghai Jiao Tong University in 2012, and PhD degree from The Chinese University of Hong Kong in 2016, respectively. Huanle’s main research interests include cloud computing, distributed systems, and machine learning. His research has been published in leading system conferences and journals, including ISCA, ASPLOS, HPCA, Eurosys, SC, ICS, SoCC, SPAA, and ToCS. Huanle is also the recipient of the 2021 ACM SoCC Best Paper Award.
Abstract: Large Language Models (LLMs) and emerging AI agents are reshaping the landscape of intelligent services by enabling autonomous reasoning, complex task execution, and multi-step decision workflows. However, their dynamic interaction patterns and heterogeneous computational demands introduce new challenges for scalable, efficient, and adaptive system support. This talk presents recent advances in cloud-native system design for fine-grained and dynamic resource management tailored to LLM- and agent-driven workloads. We explore: (1) leveraging containerization for fine-grained resource control to achieve lightweight scheduling and replication, (2) dynamically balancing workloads through adaptive resource allocation and scaling, and (3) adaptive scaling and workload orchestration mechanisms that respond to agent behaviors, fluctuating task graphs, and evolving model demands. Collaborating with industry leaders, our research bridges LLM serving and agent execution pipelines, enabling cost-effective, high-throughput, and robust cloud-native AI infrastructures. The outcomes aim to advance system-level support for LLMs, contributing to improved performance and reduced costs for cloud-native AI services.
Biography: Dr. Minxian Xu is currently an Associate Professor and PhD supervisor at Shenzhen Institutes of Advanced Technology, Chinese Academy Science (SIAT). His research interests are resource management for Cloud-Native systems and AI infrastructure. He has published 80+ research papers in prominent journals and conferences (3 ESI highly cited papers), including CSUR x3, TSC x3, TMC, TAAS, TOIT, TSUSC x5, TNSM, TCC, TGCN, TASE, TCE, ICSOC x4 and ICWS. These research work has attracted 5500+ citations (Google Scholar data). He has been the Program Chair/Program Committee for 20+ international conferences, and has given 30+ oral presentations. He is the PI/Co-PI of 20+ research projects, including grants from NSFC, Ministry of Science and Technology, CAS PIFI, Guangdong NSF, Shenzhen NSF, Alibaba and Huawei. He is also the active reviewer for 50+ journals and conferences, such as TPDS, TC, ToN, TMC, TSE, TSC, TPAMI, TKDE, and ICDCS. His PhD thesis was awarded the 2019 IEEE TCSC Outstanding PhD Dissertation Award, and he was honored to be awarded the 2023 IEEE TCSC Award for Excellence (Early Career Award), and he is among the 2023 to 2025 World's Top 2% Scientists by Stanford University. He is a Senior Member of both IEEE and CCF.
Abstract: Autonomous surface and underwater vehicles have become indispensable platforms for exploring, observing, and operating in the ocean in ways that are safer, more efficient, and frequently unattainable with conventional approaches. This presentation provides an overview of a family of marine robotic systems developed in our group – including survey AUVs, intervention vehicles with manipulators currently under development, and autonomous surface craft – and illustrates how they can be deployed in applications ranging from coastal monitoring and offshore inspection to deep-sea operations and marine archaeology. Rather than concentrating on specific algorithms, the talk will emphasise the research opportunities these platforms enable for interdisciplinary collaboration with oceanography, marine biology, geoscience, and offshore engineering, where robots function as mobile sensing and intervention laboratories. A dedicated part of the presentation will discuss key challenges in guidance, navigation, control, and state estimation that arise when these vehicles are required to operate over extended durations, coordinate in teams, work in proximity to complex structures, or interact physically with the environment. Finally, we will outline future directions in which these systems can support climate and environmental observation, contribute to safer and more sustainable offshore operations, and serve as open testbeds for novel concepts in autonomy and intelligent sensing.
Biography: Carlos Silvestre (Senior Member, IEEE) received his Licenciatura and M.Sc. in Electrical and Computer Engineering, his Ph.D. in Control Science, and his Habilitation in Electrical and Computer Engineering from Instituto Superior Técnico (IST), Lisbon, Portugal, in 1987, 1991, 2000, and 2011, respectively. He has been with the Department of Electrical and Computer Engineering at IST since 2000, where he is a Professor in Systems, Decision, and Control, currently on leave since 2012. He is currently a Professor and Head of the Department of Electrical and Computer Engineering at the University of Macau’s Faculty of Science and Technology. His research interests include linear and nonlinear control, estimation theory, hybrid systems, multi-agent control, networked control, inertial navigation, and machine learning for autonomous systems, with a focus on unmanned ocean and aerial vehicles.
Abstract: Agentic AI systems built on large language and vision-language models are beginning to plan, act, and adapt with increasing autonomy, often running on constrained devices and in semi-trusted or adversarial environments. Making such agents practical requires them to be lean and safe. This talk explores how to build that foundation at the level of the underlying models. I will outline efficiency and robustness considerations and methods towards this goal, and provide a future vision for open research questions in this area.
Biography: Xitong Gao received his MEng degree in information systems engineering in 2012, and a PhD degree in electronic engineering research in 2016, both from Imperial College London. He is currently working as an associate researcher at the Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences. His research interests include efficient inference and training and safety and privacy risks of large language models, and by extension, agentic AI models.
Abstract: In this talk, I will present linear estimation strategies for two fundamental angular
motion problems: (i) estimating the angular velocity of a rigid body with a known inertia matrix, and (ii)
estimating the inertia matrix itself when angular velocity measurements are available. Both problems exploit
attitude measurements and torque inputs, while explicitly accounting for constant external disturbances in
the dynamics. These disturbances are estimated jointly with the unknown states.
By revisiting a classic coordinate transformation, we reformulate the nonlinear dynamics into linear
time-varying models and establish their uniform complete observability. This insight makes linear Kalman
filters a natural and efficient choice for state estimation. I will illustrate the approach with numerical
simulations and share experimental results that highlight its practical applicability.
Biography: Joel Reis received the M.Sc. degree in Aerospace Engineering from the Instituto Superior Técnico, Lisbon, Portugal, in 2013, and the Ph.D. degree in Electrical and Computer Engineering from the University of Macau, Macau, in 2019. He is currently an Assistant Professor with the Faculty of Science and Technology, University of Macau. His research interests include estimation and control theory for autonomous vehicles.
Abstract: This talk presents ongoing work within the EU project REPAIR, which aims to digitally and physically reconstruct fragmented Pompeii frescoes through the integration of artificial intelligence, machine learning, and advanced robotics. As one of the project partners, our focus lies in developing robotic manipulation strategies capable of handling extremely fragile archaeological fragments with precision and care. We begin by leveraging simulation environments to study and learn grasping, picking, placing, and gentle pushing actions using a robotic arm with a soft, semi-rigid, 5-finger. These simulations allow us to explore safe interaction strategies, optimize control policies, and test puzzle-completion behaviors without risking the artifacts. After iterative refinement in simulation we transition to our own robotic testbed. In this presentation we also present the current approach used on the official robotic platform deployed on-site in Pompeii. This project demonstrates how AI-driven manipulation, when combined with domain knowledge from archaeology, can support the delicate process of cultural heritage restoration and pave the way for new human–robot collaborations in conservation.
Biography: Nuno Ferreira Duarte is a Researcher at ISR-Lisboa in Instituto Superior Técnico. He received his Ph.D (Summa cum laude) in 2023. His PhD was a Joint-Degree between Instituto Superior Técnico (IST-Lisbon) and École Polytechnique Fédérale de Lausanne (EPFL) advised by Professor José Santos-Victor and Professor Aude Billard. He received a M.Sc (Cum laude) from the University of Lisbon, IST-Lisbon. He also interned at California Institute of Technology (Caltech SURF Program) under the supervision of Professor Richard Murray. His main topics of research are Robotics and Computer Vision with an emphasis on human action understanding from neurological, psychological, and physiological behaviour. The importance non-verbal communication behaviour has on decoding human action intention. Using non-verbal cues as a communication language between humans and robots with the goal of having robots that understand actions as well as understood by humans.
